tests: pin inline-lambda type-variable params across engines

[danieljohnmorris]

Summary

nlp-engineer rerun6 reported rsrt (r:L a>n;at r 0) rows failed with cascading ILO-T002 duplicate function definition 'a' and ILO-P001 expected declaration, got RParen against v0.11.5, inferring the lambda-lifter was treating type-var a as a top-level fn name.

I couldn't reproduce against f6b3271. Every variant of the reported shape now parses, verifies, and runs identically across tree, VM, and Cranelift. src/verify.rs:170-182 already treats unaliased single lowercase letters (other than n/t/b) as type variables that lower to Ty::Unknown, and the lifter in src/parser/mod.rs always names synthetic decls __lit_N. Most likely the bug was closed in passing by one of the v0.11.4 to v0.11.5 parser/lambda fixes (#321 hof-error-parity, #324 bare-bang-nil, listlit-fnref-greedy, etc.).

Rather than just close the entry, this is a defensive regression-test-only PR pinning every variant of the reported shape so it can't regress silently in the future.

Repro (current state — no error)

main >n;
  rows=[[1.5,2],[2.5,3]];
  by-pmi=rsrt (r:L a>n;at r 0) rows;
  0

Runs clean on tree, VM, and Cranelift. Same for every variant the persona could plausibly have meant.

What's in the diff

• tests/regression_inline_lambda_typevar.rs (12 cross-engine cases):
  • identity (x:a>a;x) over map
  • polymorphic param with concrete return (x:a>n;0) over srt
  • the exact reported shape (r:L a>n;at r 0) over rsrt and srt
  • two same-named-type-var lambdas in one fn (must not collide on a)
  • the mi.ilo shape: two (r:L a>n; ...) lambdas threaded through rsrt
  • same shape lifted from a non-main top-level fn
  • other single-letter names (z, k) to pin the spec rule, not a hard-coded a
  • reserved-letter n still parses as the concrete number type (the type-var fast-path doesn't over-trigger)
  • fld with type-var accumulator ((acc:a x:a>n;...))
  • nested L (L a) so the type walker handles depth
• examples/inline-lambda-typevar.ilo so the examples harness exercises the shape on every engine and agents get an in-context working sample.

Test plan

• cargo test --release --features cranelift --test regression_inline_lambda_typevar — 12/12 pass
• cargo test --release --features cranelift --test examples_engines — passes
• cargo test --release --features cranelift (full suite) — all green
• cargo fmt --check — clean
• cargo clippy --release --features cranelift --tests -- -D warnings — clean

Follow-ups

None. The originating entry in ilo_assessment_feedback.md will be moved to the Addressed section with a link to this PR.

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ All tests successful. No failed tests found.

📢 Thoughts on this report? Let us know!