Effect lowering contracts and Effect Unit command entries

[biuld]

Scope

A standalone correctness and command-entry slice under #78, selected explicitly before the full Phase 4 gate. Preserve abstract source Effect and lower it to the existing generic closure representation.

Required behavior

• Resolve trusted Effect and primitive operation identities once and carry the contract into Core; untrusted same-named foreign data is not a compiler effect.
• Carry synonym-expanded, quantified checked WIT import schemes through THIR, Core, and linking; WIT conformance and Effect classification consume this shared authority rather than reconstructing raw HIR annotations.
• Classify effectful foreign imports before erasing the abstract type and suspend them from that evidence, without inferring effect semantics from an arbitrary single-parameter closure.
• Validate the converted Core, generated primitive declarations, and suspended import wrappers; distinguish structural checks from behavior established by execution.
• Use one entry identity selection rule for scope checks and command lowering.
• Execute selected main :: Effect Unit once via a generated zero-argument integer command entry. Normal completion returns exit code 0; traps propagate. Preserve integer main compatibility.
• State the source-visible runEffect compatibility rule truthfully as a restriction on direct resolved references, without claiming runner confinement.

Acceptance

Source tests cover inferred and annotated Effect main, aliases/re-exports, linked actions, single execution and observable order, assertion trap and suppressed later output, legacy integer entry, ambiguous/invalid entries, and untrusted same-named opaque types. Negative IR cases cover invalid lowering metadata and ordinary closures that must not be interpreted as effects. The full official-suite runtime scoreboard records recovered executions and newly exposed first blockers; modules without an executable main remain explicit.

References

• docs/feature/F-02-portable-programs.md
• docs/design/backend/fp/effects.md
• docs/design/backend/00-ir-boundaries.md
• docs/implementation/backend/effects.md
• docs/implementation/backend/wasi-platform.md
• docs/design/D-04-suite-roadmap.md

Validation

cargo fmt --all --check
PSRS_REQUIRE_WASMTIME=1 cargo test --workspace
cargo clippy --workspace --all-targets -- -D warnings
PSRS_REQUIRE_WASMTIME=1 PSRS_ORACLE=annotations cargo test -p psrs-driver --test suite -- --ignored --nocapture
git diff --check

This slice does not close the broader #78 epic or assert that all corpus modules have a runnable entry.

[biuld]

Merged in #140 as merge commit 2f47291. #78 stays open.

Evidence, 2026-10-04, Wasmtime 49.0.2, purs 0.15.16:

• PSRS_REQUIRE_WASMTIME=1 cargo test --workspace passed.
• cargo clippy --workspace --all-targets -- -D warnings passed.
• PSRS_REQUIRE_WASMTIME=1 PSRS_ORACLE=annotations cargo test -p psrs-driver --test suite -- --ignored measured L6/M7 at 124/413. All 124 exit 0. They are the files whose previous first blocker was a non-Int entry. A second run after the wrapper monotype fix measured the same 124/413 and the same blocker split. L1–L5 were unchanged: L1 904/908, L2 failing 72/72, passing resolved 270/413, L3 35/48, L4 35/50, L5 53/80.
• Remaining first blockers: missing library module 53, P10 no selected main 63, P3 86, P5 typecheck 46, P5 kind 17, P8 14, P6 6, P0 4.
• Both Required Wasmtime execution checks passed on 56a692e.

Corpus cases is now 0: the 124 bad-entry files run, and the 63 files with no selected main are not this issue.