fix: scan source tokens for alias hints, not raw text

[danieljohnmorris]

Summary

collect_hints was splitting raw source text after stripping string literals, but comments (-- ...) were left in. Any alias source word in a comment, identifier-ish artefact, or path-bearing line could fire hint: <word> -> <short> even though the user never wrote that token.

Streaming-tail persona on v0.11.1 caught this every Approach B run: hint: tail -> tl fired because the working directory was /tmp/ilo-persona-streaming-tail-rerun, and the path leaked into the batch script's header comment. Cosmetic but noisy in pipelines, and exactly the kind of false-positive that erodes trust in agent-facing hints.

Fix: re-lex inside collect_hints and scan Token::Ident only. Correct by construction — comments, strings, numbers, operators, and CLI argv all sit outside the token stream the user wrote. The == text-scan now also ignores comment contents via a small helper, so both hints behave consistently.

Repro

-- counts: tail of error log
main>n;42

Before: hint: \tail` → `tl` (canonical short form) printed to stderr on every run. After: no hint. Real alias use (tail xs` as a source identifier) still fires.

What's in the diff

• 9fc9785 fix: collect_hints re-lexes the source and walks Token::Ident. New helper strip_string_and_comment_contents keeps the == scan consistent (logos collapses = and == to one token, so that scan can't use lexer output).
• 583e04f test: five new regression tests pin the contract: alias word in comment, alias word in string, the == scan in comments, multi-word coverage across tail/filter/flatmap/length/head/append, and a positive test that real alias use still fires.
• 7e79f9c example: examples/tail-alias-comment.ilo shows an idiomatic no-hint program with tail in both a comment and a string literal. Multi-engine harness runs it on every backend.

Test plan

• cargo test --release --features cranelift — full suite green (2888 lib + all integration)
• cargo fmt --check clean
• cargo clippy --release --features cranelift -- -D warnings clean
• cargo test --test examples_engines passes (including the new example)
• Manual repro: tail in comment / string / path no longer fires; tail xs as a real call still fires.

Follow-ups

None planned. The lexer-token approach generalises to any future builtin alias added to ast::resolve_alias.

[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!