test(core): golden hash determinism guard across languages and versions

[laurigates]

Severity: MEDIUM — hash/normalization determinism

The stored anchor hash is the single value every Surface consumer compares against, yet no
test pins its exact output. The canonical token stream can change for an unchanged symbol
without anyone touching hash.rs:

• a tree-sitter grammar bump renames a node kind or reshapes the tree — grammars are
  caret-pinned in Cargo.toml (tree-sitter-rust = "0.24.2" ⇒ ^0.24.2) and only frozen by
  Cargo.lock, which Dependabot bumps on a schedule;
• a tree-sitter core bump changes traversal;
• a refactor of the canonicalization itself.

Any of these silently re-hashes every anchored span in every downstream repo: a wave of false
DIVERGED, or — worse — two spans that should differ start colliding. CI stays green because
nothing asserts the hash value.

What this adds

surf-core/tests/golden_hash.rs, public-API only:

1. Goldens — exact 12-hex hash of a representative symbol in each family (Rust, TS, TSX,
   Python, Go). Snippets include comments + non-canonical whitespace, so the goldens also
   encode the "formatting/comments ignored" guarantee.
2. Cosmetic invariants — reformatting and a consistent local rename keep the hash.
3. Logic sensitivity — an operator flip and swapped operands move the hash, and b - a
   does not collide with a - b.

Because CI already runs the suite on ubuntu-latest and macos-14, this doubles as a
cross-platform drift guard between the three release target triples — directly addressing
"is the stored hash stable across target triples / rust-version / grammar bumps?".

A failure here is the intended signal: the canonical form moved. The fix is to revert the
bump, or to ship the change deliberately (update CHANGELOG, treat as a hash-format break, tell
consumers to re-verify).

Verified: cargo test -p surf-core --test golden_hash → 3 passed.

Applies equally upstream (Connorrmcd6/surface).

🤖 Generated with Claude Code

[Connorrmcd6]

Great catch on the golden hash pinning, I didn't think of that! One minor change just to keep things tidy: could you drop cosmetic_edits_do_not_change_the_hash since it duplicates rust_quiet_on_reformat_and_rename (surf-core/tests/hash.rs:126)?

[laurigates]

Thanks! Dropped cosmetic_edits_do_not_change_the_hash in ffadf77 — you're right that it duplicated rust_quiet_on_reformat_and_rename (surf-core/tests/hash.rs:126). The pinned per-language goldens in golden_hashes_are_stable_per_language are the unique guard the new file adds. Remaining suite is green.