feat(cli): show codex rate limits in status bar

[schin-300]

Summary

Adds Codex rate-limit monitoring to the status bar. When the active provider is openai-codex, the bar shows remaining request quota and time until the rate window resets.

Example: ⚕ codex │ 12.4K/200K │ 6% │ 🎫 42 reqs (reset 15m) │ 14m

Changes

run_agent.py

• Added _codex_rate_limits, _codex_rate_limits_ts, _codex_rate_limits_interval to agent init
• Added _refresh_codex_rate_limits() — makes a HEAD request (falls back to GET /models) to the Codex API, parses x-ratelimit-* response headers, caches for 60s
• Hooked into agent loop after each API call (no-op when cache is fresh or provider isn't codex)

cli.py

• _get_status_bar_snapshot() — includes codex_rate_limits, codex_remaining_pct, codex_resets_in_min
• _build_status_bar_text() — shows N reqs (reset Xm) in medium and wide terminals
• _get_status_bar_fragments() — styled version with 🎫 prefix

tests/cli/test_cli_status_bar.py

• TestCodexRateLimits.test_snapshot_includes_codex_limits_when_present
• TestCodexRateLimits.test_snapshot_has_no_codex_limits_when_absent
• TestCodexRateLimits.test_status_bar_shows_codex_reqs_on_wide_terminal

Testing

• pytest tests/cli/test_cli_status_bar.py -v
• Manual: connect with Codex provider, verify status bar shows rate limits
• Manual: verify no display on non-Codex providers

Design Notes

• 60s cache interval keeps quota cost minimal (1 extra request per minute max)
• HEAD request preferred (no body transfer), falls back to GET if 405
• Gracefully degrades: empty dict when limits unavailable, no errors shown
• Narrow terminals (<52 cols) skip the display entirely