Wait for exact-base quality baseline artifacts before comparison

[karthiknadig]

Problem

Exact-base quality workflows can start before the corresponding baseline workflow for a freshly merged base commit has uploaded artifacts. The download action fails immediately instead of waiting, causing a false-negative PR check.

Observed on PR #515:

• base commit: ae216cad97a11eb389585edf3613a22dde4223b7
• performance run: 31542690250
• Linux benchmark itself completed successfully with healthy metrics
• comparison failed closed with Snapshot file does not exist: baseline-perf/metrics.json
• the exact-base Performance Baseline workflow completed successfully afterward
• rerunning only the failed Linux job then passed

Coverage has the same exact-base artifact dependency and can race similarly.

Scope

• wait for the exact-base performance/coverage baseline workflow and required artifact before download;
• use a bounded timeout and explicit polling interval;
• fail immediately with a clear error if the exact-base workflow completes unsuccessfully;
• fail clearly on timeout or missing required artifact;
• never fall back to a moving/latest-main baseline;
• share and unit-test the waiting logic across performance and coverage workflows.

Acceptance criteria

• a pending exact-base workflow is polled until its required artifact is available;
• successful, failed, timed-out, and missing-artifact paths are tested;
• performance and coverage workflows use the helper;
• exact-base SHA remains visible in logs/reports;
• full CI, comparator tests, coverage, performance, and CodeQL pass.

[karthiknadig]

Final quality conclusion for PR #517 at 036cfbd:

• Exact-head Copilot review: 4/4 files, 0 comments, 0 unresolved threads.
• Full CI, lint, CodeQL, Linux/Windows coverage, and all performance jobs pass.
• Live performance wait selected perf-baseline-windows-latest from exact-base run 31542508159 for ae216ca.
• Live coverage wait selected coverage-baseline-windows-latest from exact-base run 31542508086 for the same SHA.
• Performance artifacts contain 10 samples, matching inventories, and empty interpreter-timeout maps on every platform.
• Coverage is unchanged on Linux and Windows; all performance metrics remain within enforced budgets.
• Helper/unit suite: 41/41 passed, including pending workflow, successful artifact publication, failed workflow, timeout, missing/expired artifact, and input validation.
• Both workflow YAML files parse, and direct real performance/coverage artifact probes pass.

The implementation never falls back to moving main; wait/download/comparison failures remain fail-closed.