coro: add the stackless preemptive runtime core

[cpunion]

Summary

This is the integrated compiler/runtime core for #1546. It keeps ordinary Go source and standard-library calls synchronous in syntax, infers suspension transitively, and lowers the selected functions to LLVM stackless coroutines.

The implementation deliberately does not generate two complete bodies for every function. A source function has one primary body; an additional callable/dispatch representation is emitted only when a value can cross an open dynamic boundary such as any, an interface, a function value, reflection, or foreign code.

What this adds

• A whole-program effect/demand plan with canonical function identity, frozen per-site lowering facts, cache digests, archive validation, and fail-closed dynamic-call handling.
• A structured LLVM coroutine builder and physical ABI for LLVM 19–22, including frame allocation/root retention, child await, preemption safepoints, explicit panic outcomes, defer/recover cleanup, and dynamic callable dispatch.
• Compiler-injected callable metadata for C/assembly/host functions. Blocking behavior is never recovered from a raw function address; exact wrappers and call sites carry content-addressed contracts into the plan.
• One runtime scheduling model built around G, P, executor ownership, OperationID, ParkState, WaitSetRecord, result leases, select winner claims, cancellation, detach, and producer quiescence.
• A shared event-source model for timer, poll, worker, channel, semaphore, and notify waits, including lossless idle admission and bounded run slices.
• A native two-domain executor fleet with exact routed wakeups, a bounded worker pool for blocking syscalls/files, poll/deadline integration for sockets, and strong shutdown/join semantics.
• Target-neutral host-pull adapters and cross-target compilation coverage for JS/WASM, WASI, embedded, and bare-metal environments without requiring libuv or BDWGC in the scheduler core.
• Go 1.26 synchronous-source acceptance probes that compile, link, and run through LLGo for time.Sleep, time.Timer/Ticker, direct file syscalls, os.File, and loopback TCP with deadlines.

Architectural cutover

This branch has completed the planned hard cutover for the core implementation:

• no staged EnableCoro* feature surface;
• no legacy logical wait registration or Timer/Poll dual modes;
• one static source dispatcher and one native fleet path;
• one frozen physical plan for helper discovery, preflight, emission, and cache identity;
• zero-tolerance architecture gates prevent the removed protocols from returning.

The design and invariants are documented in:

• doc/llvm-coro-runtime-design.md
• doc/coro-ir-design.md
• doc/coro-async-core-contract.md
• doc/coro-callable-contract.md

Validation

Validated locally with Go 1.26.5, LLVM 19.1.7, Darwin/arm64:

• go test ./ssa -count=1 — pass (109 s)
• go test ./cl -count=1 -timeout=20m — pass (1106 s)
• go test ./internal/build -count=1 -timeout=20m — pass (582 s), including the five standard-library acceptance programs
• runtime core, allocator, doorbell, fleet, channel, keyed wait, and timer tests under -race -shuffle=on
• JS/WASM runtime adapter and channel execution
• JS/WASM, WASIp1, Linux ARM/RISC-V, and Cortex-M/bare-metal cross-compilation
• all coroutine architecture/hard-cutover gates
• all repository coroutine vet lanes

The dedicated Coroutine workflow adds LLVM 19, 20, 21, and 22 jobs, plus integration and target lanes. The regular upstream workflows are re-enabled for this integrated PR.

Scope boundaries

This PR lands the reusable core and proven vertical slices; it does not claim that every target or every GOROOT test is complete. Follow-up work remains for dynamic P sizing/work stealing, full timer GC/synctest/asynctimerchan semantics, broader FD/process/signal coverage, the ordinary JS/WASI command reactor pump, RTOS/bare-metal HAL adapters, precise moving-GC integration, and the remaining open dynamic FFI/tooling surface.

LLVM dependency

go.mod temporarily replaces github.com/xgo-dev/llvm with github.com/cpunion/llvm@bf5fb88be315 for the LLVM 22 bindings used by the matrix. The corresponding upstream proposal was xgo-dev/llvm#43; this replacement can be removed once equivalent bindings are available upstream.

Part of #1546.

[gemini-code-assist]

Caution

The consumer version of Gemini Code Assist on GitHub has been sunset. All code review activity has officially ceased.

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.

📢 Thoughts on this report? Let us know!

[cpunion]

Keeping this experimental coroutine line in cpunion/llgo for now. The upstream CI result has been recorded; development and validation will continue in the fork.