Performance optimization sweep

[Juliusolsson05]

Summary

Addresses #103.

This PR collects the first performance optimization sweep. It is intentionally scoped to high-confidence fixes from the performance research pass; larger follow-ups like per-session runtime subscriptions, push-based orchestration lifecycle, and MCP server lifecycle caching should stay separate.

Current commits cover:

• stream initial and older history loading while limiting restore concurrency
• bound performance telemetry overhead and tail reads
• reduce renderer layout churn around ghost merging, split drag, and terminal resize
• avoid per-row editor directory stats
• cache worktree activity summaries
• stream transcript inspect/read/search paths
• compact and retain ghost logs
• defer hidden Monaco code blocks inside collapsed result details
• cache worktree git probes and cap worktree status fanout
• reduce AI Workspace registry churn
• reduce orchestration wait polling work
• split feed committed render projection

Tests

• npm run test:agent-transcripts
• npm run test:ghost-fallback
• npm run test:dispatch-terminals
• npm run test:feed-render-model
• npm run test:worktree-activity

Note: full tsc remains blocked by pre-existing local package declaration / .tsc-out issues on main.