fix: optimize prepare-commit-msg hook performance for large repos

[pfleidi]

Summary

• Replace getStagedFiles with native git CLI: Swap go-git's worktree.Status() (O(all files), scans entire working tree) with git diff --cached --name-only (constant time, uses git's optimized index and fsmonitor)
• Cache staged files across sessions: Compute staged files once in filterSessionsWithNewContent and pass via contentCheckOpts struct, eliminating up to 3×N redundant calls for N sessions
• Optimize getLastPrompt: Read prompt.txt directly from the shadow branch tree instead of calling extractSessionData() which parses the full transcript (token counting, context generation, prompt extraction)
• Add benchmark harness: BenchmarkPrepareCommitMsg with 6 sub-benchmarks (Small/Medium/Large repo × 1/3 sessions) and BenchmarkGetStagedFiles for isolated measurement

Also includes the prior fix for entire explain hanging on repos with many checkpoints (commit eff1e18d).

Test plan

• mise run fmt — clean
• mise run lint — 0 issues
• mise run test:ci — all tests pass (unit + integration)
• Benchmarks run successfully at all repo sizes
• Manual test: git commit with Entire enabled on a large repo no longer hangs

🤖 Generated with Claude Code

[cursor]

PR Summary

Medium Risk
Changes git hook behavior and performance by switching staged-file detection to git diff --cached and refactoring session content checks, which could affect trailer linking decisions in edge cases or environments where git execution differs.

Overview
Performance-focused refactor of git hooks and checkpoint listing. prepare-commit-msg now detects staged files via native git diff --cached --name-only, caches that result across sessions, and threads it through a new contentCheckOpts path to avoid repeated expensive checks.

Avoids heavy transcript parsing in hot paths. getLastPrompt and entire explain’s branch checkpoint list now read prompt.txt directly from git trees (including a single cached metadata-branch tree), instead of loading/parsing full session transcripts per checkpoint.

Reduces expensive go-git diffs in list views. Temporary checkpoint filtering switches from a full diff (hasCodeChanges) to an O(1) tree-hash comparison (hasAnyChanges), and adds benchmark coverage for PrepareCommitMsg and getStagedFiles plus test updates for the new APIs.

^{Written by Cursor Bugbot for commit ea28a47. Configure here.}

[Copilot]

Pull request overview

Improves performance of Entire’s git hook and checkpoint listing paths in large repositories by avoiding expensive go-git operations and repeated work, while adding benchmarks to measure improvements.

Changes:

• Replaces staged-file detection with git diff --cached --name-only and caches staged files across session checks.
• Optimizes last-prompt retrieval by reading prompt.txt directly from git trees instead of parsing full transcripts.
• Reduces entire explain checkpoint-list overhead by reading prompts from the metadata tree and using a cheap “any changes” check.

Reviewed changes

Copilot reviewed 3 out of 3 changed files in this pull request and generated 4 comments.

File	Description
cmd/entire/cli/strategy/manual_commit_hooks.go	Caches staged files, switches staged-file detection to git CLI, and optimizes last-prompt lookup via prompt.txt in trees.
cmd/entire/cli/strategy/mid_turn_commit_test.go	Updates tests to pass staged files via the new contentCheckOpts path.
cmd/entire/cli/explain.go	Avoids slow transcript reads in list views by reading prompts from the metadata tree and using a lightweight tree-hash change check.
cmd/entire/cli/strategy/preparecommitmsg_bench_test.go	Adds benchmark harnesses for PrepareCommitMsg and staged-file detection across repo sizes/session counts.