fix: cranelift JIT helper nil-sweep batch 1 (lst, index, slc, jpth, listget)

[danieljohnmorris]

Summary

Batch 1 of the Cranelift JIT permissive-nil sweep. Companion to #259 (batch 2: recfld / recfld_name / unwrap / mget) and #254 (the JIT_RUNTIME_ERROR TLS channel itself). These helpers used to collapse type and range failures into a silent TAG_NIL return, hiding bugs that tree/VM already diagnose. After this PR, every engine surfaces the same runtime error on the same input.

Manifesto framing: fewer engine-conditional retries. An agent that hits lst xs 999 v should learn it is out of range whether the program is interpreted or JITed. Today the tree path teaches that, the Cranelift path silently passes a stale list along, the next operation fails for a different reason, and the agent has to triangulate. This closes the loop.

Repro before / after

# Before this PR
$ ilo "f>L n;lst [1,2,3] 5 99" --run-cranelift f
[1, 2, 3]   # silent nil-sweep: returns the input list, exit 0

# Tree and VM already errored
$ ilo "f>L n;lst [1,2,3] 5 99" --run-tree f
{"code":"ILO-R009","message":"lst index 5 out of range","severity":"error"}

# After this PR (all three engines)
$ ilo "f>L n;lst [1,2,3] 5 99" --run-cranelift f
{"code":"ILO-R004","message":"lst: index out of range","severity":"error"}

Same story for xs.5 on a 3-element list, lset xs -1 v, slc xs "a" "b" (non-number indices), and jpth 42 "a" (non-string args).

What's in the diff

1. jit: route lst/index/slc/jpth/listget failures through JIT_RUNTIME_ERROR (src/vm/mod.rs): per-helper edits.
   • jit_lst: error on OOB, negative idx, fractional idx, non-number idx, non-list. Diagnostic strings match tree/VM wording.
   • jit_index: error on OOB and non-list (verifier-blocked but defensive).
   • jit_slc: error on non-number indices and non-list/non-text args only. OOB clamping is the documented tree/VM contract and is left alone — inline comment notes the asymmetry.
   • jit_jpth: error on non-string args. Path miss still returns Err(...) as a Result value.
   • jit_listget: error on non-list collection and non-number idx. OOB stays TAG_NIL because it is the foreach loop-done sentinel.
   • Helper-level unit tests for jit_lst updated to assert TAG_NIL + the JIT_RUNTIME_ERROR cell.
2. test + example: cross-engine regression for jit nil-sweep batch 1: new tests/regression_jit_nil_sweep_batch1.rs (28 tests across tree / VM / Cranelift) and examples/jit-nil-sweep-batch1.ilo (positive paths exercised by the examples_engines harness). Includes negative regressions for the deliberately permissive paths (slc OOB clamp on list + text, jpth path-miss returning Err) and a stale-error-leak guard for the new helpers.
3. test: update suites that pinned the old permissive-nil contract: four pre-existing tests in eval_inline.rs, regression_dot_list_index.rs, regression_list_mutation.rs, and regression_lset_alias.rs flipped from asserting the old passthrough to asserting the new error parity.

Test plan

• cargo build --release --features cranelift
• cargo test --release --features cranelift (full suite green)
• cargo clippy --release --features cranelift --all-targets -- -D warnings
• cargo fmt --check
• New regression file: 28 tests pass across tree, VM, Cranelift
• examples_engines harness runs the new example across all three engines
• Stale-error-leak guard: successful run after an erroring one in a fresh process must not inherit the TLS error

Follow-ups

• The helpers in this batch can adopt the span-id immediate pattern (one-line iconst per call site + one extra u64 arg in the signature) when that PR lands. The span-id PR is in flight on a parallel branch; no rebase blocker either direction.
• Batches 1 and 2 together complete the explicit permissive-nil audit. If a third batch surfaces it would be helpers found by future fuzzing or by audit of opcodes that didn't appear in the original brief.

[codecov]

Codecov Report

❌ Patch coverage is 91.30435% with 4 lines in your changes missing coverage. Please review.
✅ All tests successful. No failed tests found.

Files with missing lines	Patch %	Lines
src/vm/mod.rs	91.30%	4 Missing ⚠️

📢 Thoughts on this report? Let us know!