fix(git): prevent diff stats polling process storms

[qixing-jk]

Summary

• keep diff-stat polling stable across rerenders when the active worktree path set is unchanged
• share the 30-second polling lifecycle between tree and column layouts
• coalesce concurrent git diff --shortstat requests per worktree and reuse completed results for two seconds

Root cause

useQueries() can return a new result-array reference on each render. EnsoAI rebuilt worktreesMap from that array, and TreeSidebar treated the new map reference as a polling dependency change. Because the effect fetched immediately before starting its interval, ordinary rerenders repeatedly bypassed the intended 30-second cadence and spawned short-lived Git processes.

Validation

• pnpm test - 68/68 tests passed
• pnpm build - main, preload, and renderer production builds passed
• targeted Biome check for all changed source/test files passed
• git diff --check upstream/main...HEAD passed
• temporary cross-layer stress harness passed twice: 100 equivalent rerenders and 33 concurrent renderer requests produced one real Git process and one shared result

Known repository baseline

• pnpm typecheck still reports the pre-existing errors at src/renderer/App.tsx:472 and src/renderer/App.tsx:476; neither line is changed by this PR
• repo-wide pnpm lint remains blocked by existing formatting/lint debt, while all files changed by this PR pass Biome

Risk

Diff statistics can be up to two seconds stale after a completed request. Error and empty results use the same short cache window to prevent retry storms. No production dependency or user-facing UI change is included.

[github-actions]

Claude Code is working…

I'll analyze this and get back to you.

View job run