Reduce macOS thread processing overhead

[SergeSerb2]

Summary

• project oversized non-MCP activity payloads at HTTP and WebSocket boundaries, adapting the upstream T3 Code transport optimization while preserving full persisted events and MCP payloads
• keep only the visible macOS timeline subscribed and coalesce overlapping cold timeline loads
• replace 1 ms scenery image polling with continuations
• debounce sidebar search projection work and cache project activity timestamps used during sorting
• retain useful command, file, and output summaries in native activity rows after payload projection

This reduces network transfer, JSON decoding, retained observation work, main-actor wakeups, and history-sized sidebar work on the native macOS path. The large-payload regression fixture verifies a greater than 90% reduction for its representative activity event; upstream reports roughly 50% aggregate WebSocket throughput reduction in pingdotgg/t3code#4622.

Area

• apps/mac — native macOS app
• apps/windows — Windows desktop app
• apps/mobile — iPhone companion app
• apps/server — backend server
• Shared packages (packages/contracts, packages/shared, packages/client-runtime, …) or relay (infra/relay)
• Build, CI, or release tooling
• Docs

Verification

• pnpm run verify --all passes (check + typecheck + tests; adds lint:mobile and the Swift suite)
• pnpm run test:mac passes (710 tests)
• Manually verified (describe below)

Targeted verification also passed 109 related server tests, including projection immutability, MCP passthrough, compact field retention, and the large-payload size bound.

Screenshots / Recordings

Not applicable; behavior and transport performance changes only.

Release notes

Makes long threads, timeline switching, sidebar search, and scenery loading more efficient in the native macOS app.