test(coverage): restore to ~99.6/99.0/99.9/99.9 + gate coverage in CI

[gyng]

Coverage had slid to 97.69 / 93.45 / 96.39 / 98.89 unnoticed (Vitest 2→4 remap changes in #54 plus untested-code drift) because CI ran `test:unit` but never the thresholds. This PR restores it above the original v0.0.48 baseline and makes regression impossible to miss.

What's in it

• ~400 new tests across 5 parallel clusters — editor panels (Inspector 81.1% → 99.7% stmts, ColorField/themeTokens/Outline → 100%), meters (all listed files → 100%), plugins+llm+formula, lib/core (26/38 files at flat 100), canvas + widgets misc. Suite: 2,769 → 3,151 tests.
• All test-only: zero source changes, zero coverage-ignore pragmas (standing policy). The residue is documented genuinely-unreachable arms — type-union-impossible branches, guards mirrored by `disabled` attributes, library-internal arms (downshift, lezer — the latter fuzz-verified over 60k fragments) — each justified in the agents' notes.
• CI now gates coverage: the client job runs `test:coverage`, so the vite.config ratchet fails any regressing PR.
• Thresholds re-ratcheted: statements 99→99.25, lines 99→99.5, functions 96→99.5, branches 95→98.5 (~0.4pp slack for cross-platform v8 remap jitter).
• One pre-existing flaky test hardened (MonitorSourcesEditor rescan raced a disabled button under parallel load).

Verification

Final tree: `tsc` clean, oxfmt+oxlint clean, 3,151 tests / 276 files pass, build clean, coverage gate exit 0 at the new floors. Global: 99.62 stmts / 98.97 branches / 99.91 funcs / 99.93 lines.

No product code changed → no release from this PR (per instruction).

🤖 Generated with Claude Code