[Feature] purwc: a single module compilation worker - purs-wasm orchestrate mode and nix-style cache store

[katsujukou]

What's in this PR

Separated compilation — ADR 0038

• cli-lib: common CLI machinery extracted from purs-wasm / ulib-tooling into a shared package.
• purwc: a standalone single-module worker. It compiles one module from its source and its dependencies' .pmi interfaces into three artifacts — {M}.pmi (the dependent-facing interface + optimization summary), {M}.wasm (the object), and {M}.link.json (link metadata). The .pmi is now a complete interface (no .pmo needed by dependents).
• compile-batch: a long-lived worker mode that compiles a topo-ordered work-list in one process, amortizing Binaryen init across modules.
• Orchestrator: purs-wasm build --orchestrate drives purwc per module (in dependency order) and links the results.

Content-addressed library store — ADR 0040

• Cache key: nix-style source/toolchain tag hashing, and a recursive .pmi key — a module is keyed against its dependencies' .pmi keys (content-address), not their summary hashes.
• Foreign self-merge (P2): a module's kept foreign .wat is merged into {M}.wasm, so a worker artifact is self-contained.
• caf_init reachability ([Feature] Monomorphization (unbox polymorphic container elements) #19): per-binding reference graph emitted by the worker; the orchestrator chains caf_init only for entry-reachable modules.
• The store: $PURS_WASM_STORE holds the {.pmi, .wasm, .link.json} triplet per library module, keyed by content. Write-back is partitioned by own vs library — a .spago dependency or a ulib shadow goes to the global store (shared across builds and projects), the project's own modules stay in the local _build. store; reuse is content-addressed and sound (version drift → key mismatch → recompile).
• prewarm (optional): precompile a package set's library closure into the store.
• .pmo retired (P6): the .pmi interface + the .wasm object are the artifacts; the separate optimized-MIR object is gone. The in-process whole-program path stays as an uncached differential oracle.

Fixes

• Inliner self-pollution fix (exclude the module itself from its inline summary), and the generic-dispatch size cap lowered to 8000.
• Lower: a non-functional recursive value around Control.Lazy.fix was mis-handled.
• CI: the compiler-leg e2e prebuild now builds purs-wasm explicitly (it had relied on ulib-tooling transitively pulling it in, which no longer holds since the maintainer CLI was decoupled from the orchestrator).

Cleanup & docs

• User-facing CLI messages/help no longer mention internal ADR numbers.
• ADR 0038/0039/0040 updated to reflect what shipped; new ADR 0041 (ship pre-built corefn behind a build-time content-based compatibility gate; drop ulib install) and ADR 0042 (orchestrate as the default build; trigger-based retirement of the whole-program oracle) capture the agreed next direction (Proposed).

Checklist

CI passing is enforced by the required ci-gate status check, not by a box here.
The items below are the human-judgment gates:

• Docs updated — any behaviour/feature/representation change is reflected in docs/ and relevant ADR(s).
• Tests — added/updated for this change; a bug fix includes a regression guard in the routinely-run lane (unit / e2e), not only a slow test:bin script.
• Design changes have an ADR — added or updated, with its Status set (ADR-first).
• No perf regression — for optimizer / runtime / lowering changes, benchmarks compared (node bench/run.mjs <dir>; do not overwrite bench/snapshots/baseline.json). N/A otherwise.
• Runtime GC-type / ABI / canonicalization — changes to the value-type substrate or the host/runtime ABI keep cross-module type canonicalization intact. N/A otherwise.