feat: add repo staleness detection and refresh across MCP, API, and CLI

[eddiefleurent]

Summary

• Indexed repositories were treated as immutable forever — no git pull, no commit comparison, no TTL. The only way to get fresh data was manually deleting index metadata via delete_repo_metadata and re-triggering indexing.
• Added lightweight staleness detection (git fetch + SHA comparison) and an explicit refresh flow (git pull + re-index) with feature parity across all three interfaces (MCP, REST API, CLI).
• The UX approach is non-blocking: code_qa appends a tasteful freshness warning when repos are outdated, nudging the user to call refresh_repo — queries still work against the existing index.

Changes

File	Change
fastcode/loader.py	New methods: get_head_commit(), check_for_updates() (git fetch + SHA compare), pull_updates() (git pull); fixed hexsha[:8] → hexsha to prevent stale false-positives
fastcode/vector_store.py	save() now persists indexed_commit SHA in metadata pickle; new get_indexed_commit() reader
fastcode/main.py	index_repository() and load_multiple_repositories() pass commit SHA to save(); new check_repo_for_updates() and refresh_repository() orchestration methods
mcp_server.py	_ensure_repos_ready() checks freshness on indexed repos and collects warnings; code_qa appends "Repository freshness" section when stale; new tools: check_repo_freshness, refresh_repo
api.py	New endpoints: POST /check-repo-freshness, POST /refresh-repo
web_app.py	New endpoints: POST /api/check-repo-freshness, POST /api/refresh-repo
main.py (CLI)	New commands: check-freshness --repos <names>, refresh <repo_name>

Test plan

• Index a repo via MCP code_qa, then push a commit to the remote — subsequent code_qa call should show a freshness warning in the response
• Call check_repo_freshness tool — should report the repo as OUTDATED with commit SHAs
• Call refresh_repo tool — should pull new commits and re-index, reporting old/new commit
• Call refresh_repo again immediately — reports already up-to-date, reindexed: false, no redundant re-index
• Verify POST /check-repo-freshness and POST /refresh-repo return correct JSON via REST API
• Verify check-freshness and refresh CLI commands work end-to-end
• Verify old indexes (without indexed_commit in metadata) still load and work — staleness check reports indexed_commit: null and stale: false rather than crashing

Bug found and fixed during testing

loader.py was storing hexsha[:8] (short SHA) at index time, but all comparison code uses full hexsha. This caused indexed_commit != current_commit to always be true, reporting every repo as stale even when fully up-to-date. Fixed in the final commit by removing the [:8] slice.

Tested on

Self-hosted FastCode instance running as systemd services. Confirmed against a live repo:

POST /check-repo-freshness  (after fresh index)
{
  "stale": false,
  "indexed_commit": "2e929df112e7746af2fdfb5b6491f3c19b24015a",
  "current_commit": "2e929df112e7746af2fdfb5b6491f3c19b24015a",
  "remote_commit": "2e929df112e7746af2fdfb5b6491f3c19b24015a",
  "has_remote_updates": false
}