engine: detect self-indexing in subscript polarity analyzer

[bpowers]

Summary

The Expr2::Subscript arm in analyze_expr_polarity_with_context
(added in #491) returned current_polarity whenever the array's
identifier matched from_var, without inspecting the index
expressions. For self-indexing patterns like arr[INT(arr[i])],
the relationship between arr and the subscripted value is
non-monotone -- changing arr shifts both the lookup target AND
the index. The arm now walks index expressions via
expr_references_var and returns Unknown when any index
references from_var.

The dominant cases (SUM(arr[*]), arr[Region], indices that
reference other variables) are unaffected -- those indices do not
reference from_var, so the existing behavior of returning
current_polarity is preserved.

Why now

Surfaced by a code review on #491 (the array reducer polarity PR)
after that PR merged. The reviewer flagged it as non-blocking but
worth generalizing because the arm's docstring claimed the case was
fully handled. This PR follows up to make the docstring's claim
true.

Test plan

• New unit tests covering: literal index, wildcard, var-not-from_var,
  var-from_var (NEW), nested self-indexing (NEW), other-array.
• Existing polarity / cross-dim tests pass unchanged.
• cargo clippy --workspace --all-targets -- -D warnings clean
• cargo fmt --check clean

[bpowers]

@codex review

[claude]

Review

I reviewed the diff and don't see any blocking issues.

The change is a strict refinement of the prior Subscript arm: when ident == from_var, the indices are now walked for any reference to from_var, returning Unknown if found and current_polarity otherwise. The behavior for non-self-indexing cases (literal indices, wildcards, ranges over constants, indices over other variables) is preserved, so the reducer monotonicity guarantee still carries through for the dominant cases (SUM(arr[*]), arr[Region], etc.).

The inline index match correctly mirrors how expr_references_var (lines 440-450 of the same file) walks IndexExpr2 variants — Expr and Range carry Expr2 and need recursion; Wildcard/StarRange/DimPosition carry no Expr2 so they correctly short-circuit to false. Canonicalization is consistent: expr_references_var checks ident == var || &normalize_module_ref(ident) == var for Var nodes, matching the outer arm's check.

Test coverage exercises literal, wildcard, other-var, direct arr[arr], nested arr[INT(arr[i])], and the other != from_var baseline. A Range-with-from_var case isn't covered, but the implementation is straightforward and the existing coverage demonstrates the contract.

Overall correctness: correct. The patch addresses the docstring-claim gap without altering any pre-existing behavior outside the self-indexing case.

[chatgpt-codex-connector]

Codex Review: Didn't find any major issues. Keep them coming!

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[codecov]

Codecov Report

❌ Patch coverage is 75.00000% with 2 lines in your changes missing coverage. Please review.
✅ Project coverage is 82.43%. Comparing base (04614fe) to head (e94ea5d).

Files with missing lines	Patch %	Lines
src/simlin-engine/src/ltm/polarity.rs	75.00%	2 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##             main     #499      +/-   ##
==========================================
- Coverage   82.44%   82.43%   -0.01%     
==========================================
  Files         243      243              
  Lines       61887    61893       +6     
==========================================
+ Hits        51021    51022       +1     
- Misses      10866    10871       +5     

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.
• 📦 JS Bundle Analysis: Save yourself from yourself by tracking and limiting bundle sizes in JS merges.