docs(hooks): generate the hook event × tool matrix from the hooks factories

[dyoshikawa]

Summary

The Hook event × tool matrix in docs/reference/file-formats.md was hand-maintained and had drifted badly:

• Missing columns for twelve hook-capable targets (Pi Coding Agent, Amp, Hermes Agent, Reasonix, Qwen Code, Vibe, Junie, Grok CLI, Kimi Code, the Kiro CLI/IDE variants, and the Claude Code / Antigravity plugin variants).
• Still carried a Gemini CLI column, which is no longer a rulesync target at all.

This PR takes the issue's "generate it so it cannot go stale" option:

• New scripts/hook-events-table.ts renders the table between HOOK_EVENTS_MATRIX markers from each hooks factory's supportedEvents in toolHooksFactories — the same source of truth the generator's skip-warnings come from. Columns follow TOOL_DISPLAY order (like the other generated tables) and rows follow canonical HOOK_EVENTS order, restricted to events at least one target supports. A hooks target missing from TOOL_DISPLAY throws.
• scripts/generate-docs-content.ts applies the renderer (and oxfmt) to file-formats.md before embedding the docs, so a stale committed table makes the existing check:docs-content drift gate fail — no new package script and no package.json change needed.
• The regenerated table now has 28 tool columns × 48 event rows; I machine-verified every cell against the *_HOOK_EVENTS constants (0 mismatches).
• The hooks prose PascalCase-translation list now names Qwen Code instead of the removed Gemini CLI target.

Testing

• pnpm cicheck green (generation is idempotent; second run produces no diff).
• Cell-by-cell verification script against toolHooksFactories reported checked rows: 48 cols: 28 errors: 0.

Closes #2522

🤖 Generated with Claude Code

[dyoshikawa]

Status: review complete and ready to merge, but intentionally left for a manual merge.

• Two review rounds (code + security) finished with zero mid-or-above findings; the round-1 findings (drift-gate hole, dead guard, oxfmt plumbing, DRY, tests) were all addressed in 76fad24.
• CI is fully green on the head commit (all 9 checks including Windows/macOS E2E).
• This PR touches package.json (one line in scripts: check:docs-content now also diffs docs/reference/file-formats.md), and the autonomous batch flow does not auto-merge PRs that edit package.json per its safety policy.

One optional follow-up noted by review (not blocking): generate-supported-tools-tables.ts still uses the pnpm exec oxfmt invocation without the Windows shell handling; unifying it with runOxfmt could be a tiny future cleanup.

[dyoshikawa]

@dyoshikawa Thank you!