test(tar-xz): close 7 coverage partials with surgical v8 ignores

[oorabona]

Summary

Close 7 coverage partials in packages/tar-xz/ via surgical v8 ignore start/stop wraps. Each wrap discloses its category in the rationale comment per the project convention authorized 2026-04-30: WRAP is acceptable for branches that are EITHER physically unreachable via the public API OR reachable-but-negative-ROI (fixture cost > catch value).

Wraps applied (7)

File	Lines	Category	Branch
file.ts	~402	negative-ROI	writeFileEntryWin32 mtime=0 false branch (Windows-only, fixture cost > value)
file.ts	~543	negative-ROI	extractFile uncommon entry types (CHARDEV/BLOCKDEV/FIFO/CONTIGUOUS — no public API produces them)
format.ts	~98	unreachable	writeString TS noUncheckedIndexedAccess guard (bytes[i] always defined within bound)
format.ts	~170	unreachable	parseHeader header[156] guard (after 512-byte validation)
format.ts	~180	unreachable	parseHeader empty-string predicate via extracted isEmptyChar const (String.fromCharCode always returns 1-char)
checksum.ts	~32	unreachable	calculateChecksum header[i] guard (after 512-byte validation)
tar-parser.ts	~348, ~376	unreachable	parseTar state-machine invariant (phase === 'PADDING' always true here) — surgical 2-pragma split

Senior pre-push review

Two-pass senior Opus review: first pass flagged 3 over-suppressed wraps that masked EXECUTED code (if-true bodies, ternary's tested arm). Implementer applied 3 fixes per verbatim recipe — narrowed to surgical scope, extracted const for ternary case, split wide wrap into two 1-line pragmas. Re-pass senior: SAFE-TO-PUSH, all wraps verified minimal-scope and correctly classified.

Honest disclosure

One NEW partial is now exposed at format.ts:186 — the ternary's === '\0' arm (legacy null typeflag, header[156] === 0). No existing test exercises this case. The previous wide wrap suppressed it silently; this PR narrows the wrap so the partial becomes visible. A 1-test follow-up could close it cleanly via a hand-crafted 512-byte header buffer with header[156] = 0. Tracked for PR-θ2.

Net coverage delta

• Branch partials in scope: 7 → 1 (the new format.ts:186 exposure).
• Total packages/tar-xz/ partials: ~13 → ~7 (6 remaining TEST candidates from prior analysis + the new exposure).
• No behavior change. Diff is +29 -4.

Test plan

• CI passes (lint, typecheck, tests, smoke matrix)
• Codecov reports checksum.ts, tar-parser.ts at 100% lines/branches; format.ts shows the :186 partial honestly; file.ts partial count unchanged for the 2 wrap targets
• No regressions on packages/tar-xz/test/ (208 tests still pass)