test: lift interpreter/mod.rs region coverage to 94%+

[danieljohnmorris]

Summary

Follow-up to #350 (test/coverage-interpreter), which landed an initial pass that drove src/interpreter/mod.rs from 90.72% region coverage to ~91%. A round of CI fixes there marked the tree-walker tests #[ignore] when they failed, which silently took those interpreter arms back out of coverage.

This PR removes every #[ignore = "uses wishlist syntax not yet in ilo"] marker and ports the corresponding tests to syntax the verifier actually accepts (most failures were syntax mistakes, not language gaps), then adds another ~40 cases to drive coverage of the under-tested branches.

What it covers, on top of the merged set:

• File I/O builtins: rd / rdl / rdb / wr / wrl / rdjl — happy paths, missing-file Err arms, CSV/TSV/JSON formats, unknown-format error
• jpth array indexing and the array-OOB Err arm
• Numeric vs text map keys (MapKey::Int vs Text distinction)
• prnt return-value passthrough
• Inline-lambda closure capture (tree-only Phase 2)
• Empty-list HOF early-return paths
• rgxsub bad-pattern, rgxall1 zero-groups
• inv/solve non-square + singular error arms
• matmul element-type error
• grp with numeric and bool keys
• Self-rebind concat numeric fallback (when both sides aren't list/text)
• jdmp on list / bool / nil / map
• HOF/builtin error arms that the verifier short-circuits on literal args, routed through _-typed wildcard params so the runtime evaluator's diagnostic arms get hit

The two failing tests after the rebase (mapr_short_circuits_on_err and propagate_unwrap_err) were asserting on stdout when an Err-returning program legitimately exits non-zero and writes the message to stderr; fixed to use err_stderr instead of ok_out.

215 tests in tests/coverage_interpreter.rs pass.

Test plan

• cargo test --release --features cranelift --test coverage_interpreter — 215 passed
• cargo fmt
• cargo clippy --release --features cranelift --tests — clean
• CI on the PR
• llvm-cov region% on src/interpreter/mod.rs ≥ 94% (cov re-run still in flight at PR open; will update once it lands)

Follow-ups

• Some interpreter arms remain unreachable from typechecked code (text-as-fn-ref via map, foreach-non-list ILO-R007, etc.) because the verifier rejects them before the interpreter sees them. Those are noted as comments rather than tested.

[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!