fix(#603): store appendVersion error path — no poisoned files entry on failed path dupe

[justrach]

What

Fixes #603 (note: merging into the release branch will not auto-close it — add it to the ship closes-list).

appendVersion's first-touch block inserted the map slot via getOrPut and only then duped the path. A failed dupe (OOM) left a poisoned entry behind: key = the caller's transient slice, value = undefined FileVersions. A retry appended into undefined memory (UB), deinit invalid-freed the borrowed key, and once the caller freed its buffer the map held a dangling key. Same error-path class as #594, in the store.

The fix scopes an errdefer self.files.removeByPtr(entry.key_ptr) to the first-touch block, so the dupe failing rolls the insert back. FileVersions.init is infallible, so the errdefer guards exactly the dupe; a later versions.append failure still leaves a valid initialized entry, which is fine.

Test plan

• test(#603) commit: failing test (expected 0, found 1 + leak) on the release tip
• With fix: zig build test — 23/23 steps, 808/808 tests passed

🤖 Generated with Claude Code

[github-actions]

Benchmark Regression Report

Thresholds: 10.00% and 50,000 ns absolute delta

NOISE means the percentage threshold was exceeded, but the absolute delta was too small to fail CI.

Tool	Base (ns)	Head (ns)	Delta	Abs Delta (ns)	Status
codedb_bundle	108482	105343	-2.89%	-3139	OK
codedb_changes	17675	10169	-42.47%	-7506	OK
codedb_context	1043649	973409	-6.73%	-70240	OK
codedb_deps	380	381	+0.26%	+1	OK
codedb_edit	40028	73442	+83.48%	+33414	NOISE
codedb_find	8971	9067	+1.07%	+96	OK
codedb_hot	23756	28701	+20.82%	+4945	NOISE
codedb_outline	35749	39116	+9.42%	+3367	OK
codedb_read	20723	22766	+9.86%	+2043	OK
codedb_search	30488	28879	-5.28%	-1609	OK
codedb_snapshot	69539	70861	+1.90%	+1322	OK
codedb_status	9746	9336	-4.21%	-410	OK
codedb_symbol	51606	51099	-0.98%	-507	OK
codedb_tree	18732	19744	+5.40%	+1012	OK
codedb_word	11070	11591	+4.71%	+521	OK