Fix Windows packaging follow-ups and harden GET_BLOCK dedup guard

[Copilot]

Follow-up fixes for merged PRs #697 (Windows release pipeline) and the reverted/redesigned PR #698 (GET_BLOCK burst-guard). Addresses GMPDIR hint breakage, inconsistent ZIP layout, config glob gaps, and dedup policy correctness for the 100ms rapid-burst guard design.

Windows packaging (release-windows.yml, package-windows-prime-cpu-cross.sh)

• GMPDIR: FindGMP.cmake reads $ENV{GMPDIR}, not a CMake cache var. Moved from -DGMPDIR=... to a step-level env: block so the hint actually reaches find_path/find_library.
• ZIP layout: Compress-Archive -Path "$stage\*" produced a flat archive. Changed to Compress-Archive -Path $stage so MSVC ZIPs have the same top-level NexusMiner-windows/ folder as MinGW cross-builds.
• Config globs: Both the workflow and package-windows-prime-cpu-cross.sh previously copied only *.config. Updated to iterate both *.config and *.conf, silently skipping whichever glob is empty.
• CMakeLists.txt warning updated to say "set the GMPDIR environment variable" instead of the ambiguous "provide GMPDIR".

GET_BLOCK dedup policy (solo.cpp, get_block_reason.hpp)

Updated dedup bypass policy to match the 100ms rapid-burst guard design on MINER:

• HEALTH_NO_TEMPLATE added to should_bypass_all_dedup(): When there is genuinely no valid template, the rapid-burst guard is counterproductive — the health timer must always make progress. HEALTH_NO_TEMPLATE now bypasses all dedup guards alongside RECOVERY_FORCED, RECOVERY_TIMER, and BLOCK_ACCEPTED.
• Pending in-flight guard restored: !should_bypass_all_dedup(reason) && kept in solo.cpp so bypass_all reasons (RECOVERY_FORCED, RECOVERY_TIMER, BLOCK_ACCEPTED, HEALTH_NO_TEMPLATE) are exempt from the in-flight guard, allowing degraded-mode retries to always make progress.
• Jitter values: Using 100–250ms (matching the 100ms rapid-burst guard design).

New helper and tests

• Added should_bypass_height_state_after_cooldown() as an explicitly-named alias for should_bypass_all_dedup() to clarify scope. Includes a TODO for future divergence when get_block_interval_ms is wired to runtime guards.
• get_block_interval_ms documented as parsed-but-not-wired to runtime burst guard constants.
• Worker_manager burst-suppression log upgraded debug → info.
• Three new test suites (Tests 18–20): rapid-burst guard exemption for bypass_all reasons, pending in-flight guard policy (bypass_all exempt; non-bypass suppressed until 4s timeout), and alias correctness across all defined GetBlockReason values.

Original prompt

Context

PR #697 (Add Windows release pipeline and MinGW cross-compilation path with PRIME backend/toolchain fixes) and PR #698 (Throttle miner GET_BLOCK recovery storms) have both merged into MINER.

PR #698 correctly introduced a universal 2-second miner-side GET_BLOCK cooldown, made HEALTH_NO_TEMPLATE stop bypassing all dedup, preserved cooldown across template-state resets, deferred malformed/empty template recovery through Worker_manager, and added PUSH→NEW_ROUND suppression when recent PUSH implies BLOCK_DATA is in transit.

The remaining follow-up work should harden correctness and operator ergonomics without undoing the storm fix. Favor correctness over convenience. Avoid broad rewrites; make targeted, testable improvements.

Relevant production screenshots/logs that motivated this work:

Goals

1. Fix Windows release packaging follow-ups from merged PR #697

Address the review comments left on PR #697:

1. .github/workflows/release-windows.yml passes -DGMPDIR=..., but cmake/modules/FindGMP.cmake consumes GMPDIR as an environment variable ($ENV{GMPDIR}), not as a CMake cache variable. Fix the workflow so the fallback hint actually works.

   • Prefer setting GMPDIR in the Configure step environment or PowerShell before cmake runs.
   • Alternatively pass -DGMP_INCLUDES=... -DGMP_LIBRARIES=... if that is more consistent with the existing FindGMP module.
   • Update the root CMakeLists.txt warning that currently says “provide GMPDIR” so it explicitly says to set the GMPDIR environment variable or pass the actual include/library cache variables.

2. Align Windows ZIP layout between MSVC and MinGW cross-packaging.

   • The cross-packaging script zips a top-level NexusMiner-windows/ folder.
   • The MSVC workflow currently compresses $stage\*, producing a flat ZIP.
   • Change MSVC packaging to compress the stage directory itself so the ZIP contains a top-level NexusMiner-windows/ directory.

3. Packaging config completeness.

   • PR Add Windows release pipeline and MinGW cross-compilation path with PRIME backend/toolchain fixes #697 root install supports both *.conf and *.config during migration.
   • The Windows packaging paths only copy configs/*.config.
   • Update both MSVC workflow and scripts/package-windows-prime-cpu-cross.sh to include both *.config and *.conf when present, without failing if one glob has no matches.

2. Harden PR #698 GET_BLOCK cooldown / in-flight correctness

PR #698 intentionally fixed storms, but it still has a follow-up risk: forced reasons can bypass pending in-flight state after the 2-second cooldown even if the pending request has not timed out yet.

Current behavior to inspect:

• GetBlockDedupGuard enforces 2s cooldown.
• should_bypass_all_dedup() currently returns true for RECOVERY_FORCED, RECOVERY_TIMER, and BLOCK_ACCEPTED.
• Solo::get_work() skips the pending in-flight guard for reasons where should_bypass_all_dedup(reason) is true.
• If pending timeout is longer than 2s, RECOVERY_TIMER / RECOVERY_FORCED can potentially send a second GET_BLOCK between cooldown expiry and pending timeout.

Implement a more precise policy:

1. HEALTH_NO_TEMPLATE must continue to respect cooldown and pending in-flight suppression.
2. RECOVERY_TIMER should normally respect pending in-flight suppression until the pending request times out.
3. RECOVERY_FORCED should not blindly bypass pending in-flight if the existing pending request is still healthy and for the same-or-higher height.
4. BLOCK_ACCEPTED is special: the previous template is spent, but it still should not hammer AutoCoolDown. Keep the 2s cooldown, and only bypass stale height state after cooldown. If a pending GET_BLOCK already exists for the replacement height, suppress duplicate unless it has timed out.

A good implementation approach is to introduce a clearer policy helper instead of overloading should_bypass_all_dedup() for both height-state and in-flight behavior. For example:

• Keep should_bypass_height_dedup(reason) for height guard bypass.
• Rename or supplement should_bypass_all_dedup() with more explicit helpers such as:
  • should_bypass_height_state_after_cooldown(reason)
  • should_bypass_pending_inflight(reason) only when the existing pending request has timed out or is known stale.

Do not break existing semantics accidentally. Add tests for:

• RECOVERY_TIMER after cooldown but before pending timeout is suppressed by pending in-flight.
• RECOVERY_TIMER after pending timeout is allowed.
• RECOVERY_FORCED does not duplicate a healthy pending request for the same-or-higher height.
• HEALTH_NO_TEMPLATE respects pending in-flight and cooldown.
• BLOCK_ACCEPTED respects cooldown and does not duplicate a healthy pending replacement request.

3. Wire or clarify get_block_interval_ms

PR #698 changed config defaults and parser minimum for get_block_interval_ms, but runtime cooldowns are hard-coded in:

• `GetBlockDedupGuard::...

This pull request was created from Copilot chat.

[NamecoinGithub]

@copilot resolve the merge conflicts in this pull request

[Copilot]

@copilot resolve the merge conflicts in this pull request

Resolved in 4823f27. The MINER branch had reverted PR #698 (replacing the 2-second cooldown design with a 100ms rapid-burst guard), causing conflicts in 4 files. Key resolutions:

• get_block_reason.hpp: Added HEALTH_NO_TEMPLATE to should_bypass_all_dedup() to match MINER's burst-guard design (health timer must never be blocked when no template exists)
• solo.cpp: Restored !should_bypass_all_dedup(reason) && in the pending in-flight guard (bypass_all reasons — RECOVERY_FORCED, RECOVERY_TIMER, BLOCK_ACCEPTED, HEALTH_NO_TEMPLATE — are exempt)
• worker_manager.cpp: Used MINER's 100/250ms jitter values (the 2100ms floor was only needed relative to the 2s cooldown, which no longer exists); updated log text from "cooldown-suppressed" to "burst-suppressed" while keeping info level
• Tests 18–20: Rewrote to reflect the 100ms burst-guard semantics (SUPPRESS_RAPID_BURST instead of SUPPRESS_COOLDOWN; bypass_all reasons are exempt from the burst guard, not suppressed)

All 44 tests pass.