…alization

…lifecycle improvements

- Add edit, find, file_change_metadata tool handlers with file_read_ledger
- Wire last_query_usage through turn lifecycle stages for accurate context bar
- Improve ACP event routing: UsageUpdate with devo/turnUsage meta, events.rs split
- Refactor tool registry_plan and router for new handler structure
- Fix session actor state, snapshots, and resume persistence

…implification, and TUI display fixes

- provider: filter empty assistant messages across Anthropic, OpenAI
  chat completions, and OpenAI responses APIs
- apply_patch: use diffy::create_patch for correct hunk headers
- edit: remove overly strict file-read ledger checks
- finalize: filter zero usage values from terminal_usages
- queued_input_id tracking for pending input items
- TUI: add read_display for path relativization, fix abandoned
  preparing tool calls, fix PendingCellList double prefix, add
  Search/ListFiles path normalization, misc event/routing fixes

'edit' and 'write' shared the same match arm in pretty_tool_call_summary,
causing edit tool calls to display as 'Running Write ...' instead of
'Running Edit ...'.

…for correct resume context

…g row cleanup

…singleflight/prewarm

…popup path truncation and stale tests

- Add unified_exec tool with background process support and subagent coordination
- Add code search tool with singleflight/prewarm and tool_search handler
- Add task_id field to SpawnAgentResult for task tracking
- Remove built-in deep research workflow, add deferred tool loading
- TUI: center-truncate file paths in reference popup, remove skill_popup
- TUI: fix chat_composer refactor, update read display to show full relative paths
- Fix stale tests: 400 error message classification, explored block rendering, spawn agent result JSON
- Update specs for agent architecture, skills, tools, and traceability