feat(gateway): expand runtime footer telemetry

[aerbaser]

Summary

• expand the gateway runtime footer with configurable fields for full context usage, account limit windows, token totals, cost, and cwd
• fetch provider account-limit telemetry with bounded timeout + cache so final replies are not blocked
• preserve the actual chat route in turn metadata so footer telemetry is not polluted by auxiliary/compression model routes
• make Codex usage lookup resilient when runtime credentials resolve outside the legacy singleton token store
• update runtime-footer docs and slash-command references

Notes

The feature stays opt-in behind display.runtime_footer.enabled. Account-limit fields (session_limit, weekly_limit) are skipped silently when telemetry is unavailable. Provider usage fetches are cached and time-bounded by usage_cache_seconds / usage_timeout_seconds.

Tests

• python -m pytest tests/agent/test_turn_finalizer_chat_metadata.py tests/gateway/test_runtime_footer.py tests/test_account_usage.py -q

[alt-glitch]

This was generated by AI during triage.

Related: this is the broadest open implementation of the runtime-footer expansion umbrella #19922 (full context usage, account limits, token totals, cost, cwd). It overlaps the open footer-usage cluster — issues #35427 / #52288 / #27536 / #26877 and competing PRs #52600 / #49300 / #49026. Not a duplicate; flagging the cluster so a maintainer can pick the canonical footer-expansion PR.

[teknium1]

Thanks for the comprehensive footer expansion. The underlying feature remains absent from current main: gateway/runtime_footer.py:105-118 and the call at gateway/run.py:11827-11834 currently support only model, context percentage, and cwd.

Problems

• gateway/runtime_footer.py:364 calls future.result() synchronously, but the footer is built from the async gateway final-send path (gateway/run.py:10529 in this PR). A slow provider request blocks the gateway event loop. Current /usage avoids this with await asyncio.to_thread(...) at gateway/slash_commands.py:3981-3992.
• gateway/runtime_footer.py:80 changes enabled footers with no explicit fields from the legacy three fields to account-limit lookups. Gateway loads raw YAML (gateway/run.py:2315-2358), so existing opt-in footer users acquire new requests and output without selecting those fields.
• The Codex credential portion is superseded by 130e2337c (agent/account_usage.py:439-494), which preserves pool support while guarding against wrong-account fallback.

Suggested changes

• Make account telemetry refresh non-blocking and render cached data immediately.
• Keep legacy implicit fields; require explicit account/token/cost field selection.
• Salvage the footer work without restoring the superseded Codex path.

Automated hermes-sweeper review.

[teknium1]

build_footer_line() is invoked synchronously from the async gateway final-send path, so future.result() stalls the entire gateway event loop until timeout. Please make this refresh non-blocking (cache-only render plus background refresh, or an async bounded await); /usage already uses await asyncio.to_thread(...) for this reason.

[teknium1]

This changes existing enabled footers that omit fields into account-usage fetches. Gateway reads raw user YAML, so runtime_footer: {enabled: true} does not receive a migration/default merge. Preserve the legacy implicit fields and require explicit selection of account-limit/token/cost fields.

[teknium1]

This account-id resilience is now covered by mainline 130e2337c through _resolve_codex_usage_credentials, including guards against incorrect pool-account fallback. Do not restore a parallel legacy path during salvage; retain the current resolver and its tests.