feat: add async dogfood progress polling

[douglasmonsky]

Summary

• add MCP async dogfood lifecycle tools: usage_dogfood_start, usage_dogfood_status, usage_dogfood_result
• add staged progress callbacks to the existing dogfood report builder so polling sees percent/stage/cache-key updates while the job runs
• track codex-usage-tracker-async-job-status-v1 in JSON contract docs/registry
• update MCP docs and both source/packaged skills to prefer the polling flow for maintainer dogfood checks

Real local evidence

• usage_dogfood_start(refresh=False, evidence_limit=3, privacy_mode="strict", write_markdown=False) completed against local data in about 14s
• observed polling transitions: start at 1%, hypothesis stage at 40%, action brief/cache keys at 70%, completed at 100%

Validation

• PYTHONPATH=src .venv/bin/python -m pytest -q --tb=short
• .venv/bin/python -m ruff check .
• .venv/bin/python -m compileall src
• python3 scripts/check_release.py
• git diff --check

Notes

• Jobs are in-process and bounded; they are cleared when the MCP server restarts.
• No SQLite schema or privacy-mode behavior changes.