feat(tools): per-write checkpoint + change card for edit & search_replace tools (B3a v2-8b, epic #1375) - #51

What this PR does

Companion to #50 (v2-8a, apply_diff wiring). Threads the per-step change card through the remaining per-write file tools:

• EditFileTool, WriteToFileTool: the per-write checkpoint write now threads diffStats, the verbatim approval diff, and autoApproved (B3a) so checkpointSave can force the compact change card.
• EditTool, SearchReplaceTool: same threading for the edit / search_replace tools.
• New editSearchReplaceTool.changeCard.spec.ts: shared change-card threading coverage for the edit/search_replace pair.
• ApplyPatchTool: awaits the patch checkpoint (no more fire-and-forget .catch(() => {})), threads per-file diffStats/diff/autoApproved into the whole-patch journal write, and retains approval diff/stats per change.

This completes the B3a tool-wiring slice of epic Zoo-Code-Org#1375.

Line count / split rationale

789 changed lines (755+/34-) across 11 files — above the 400 soft design target, below the 1000 hard cap, and close to the ~750 pre-estimate planned for this unit (plan B10); the growth over the estimate is the added per-branch auto-approval threading tests. The five tools share one wiring shape (checkpoint write + change-card threading) and the specs are the tests of that wiring; splitting by tool would fragment one inseparable behaviour into five PRs with duplicated setup.

Not included here (later v2 PRs)

• Settings/UI round trip (CheckpointSettings, SettingsView, locales) — v2-9.
• Rollback core + service — v2-10 / v2-11.
• Rollback UI + remaining epic work — v2-12..v2-15.

Local gates

• Vitest (6 touched suites): 148 passed, 5 skipped.
• ESLint --prune-suppressions --max-warnings=0 on all 11 files: clean, suppression counts unchanged.
• Prettier: clean.
• Stryker diff gate (base 60c76ff13): 84 mutants across the 5 changed files — 82 Killed, 2 Ignored (the two documented Stryker disable next-line StringLiteral exclusions in WriteToFileTool.ts), 0 Survived, 0 NoCoverage.

Rebase note (round 23)

Rebased onto the current v2-8a head 2f6cb0acb (was 60c76ff13). This PR's diff content is byte-identical across the rebase, and the Stryker diff gate mutates only changed diff lines — the outcome above therefore stands; no re-run required.