feat(gateway): add opt-in tokens_in/tokens_out/effort runtime footer fields

[vanhci]

What does this PR do?

Extends the gateway runtime footer (display.runtime_footer.fields) with three opt-in metadata fields: tokens_in, tokens_out, and effort. Also improves context_pct rendering and gives Feishu a card-style footer.

Related Issue

• Feature umbrella: [Feature]: Expand display.runtime_footer to include all variables from /usage #19922 (expand runtime_footer with all /usage variables)
• Token footer: feat(gateway): add per-turn token usage footer to gateway responses #17592 (per-turn token usage — auto-appended style)
• Reasoning effort: feat(gateway): show reasoning effort in runtime footer #64526 (effort-only)
• Closed duplicate: Add input/output token counts to gateway runtime footer #28978

This PR complements those: it exposes tokens/effort as opt-in footer fields (not auto-appended), and is the only one covering per-platform card rendering for Feishu.

Type of Change

• ✨ New feature (non-breaking change that adds functionality)

Changes Made

• gateway/runtime_footer.py:
  • _fmt_tokens(n) — compact token formatting (15932 -> 15.9k, 28 -> 28)
  • format_runtime_footer / build_footer_line accept tokens_in, tokens_out, effort; new field branches render in 15.9k, out 1.2k, effort:high
  • context_pct now renders as ctx 12% (cc-connect style)
• gateway/run.py:
  • passes tokens_in (last_prompt_tokens), tokens_out (output_tokens), effort (agent.reasoning_effort) into the footer builder
  • Feishu renders the footer as a divider + quote block (--- + > ) so it reads as a metadata card instead of inline text

How to Test

1. Set display.runtime_footer.enabled: true and fields: [model, context_pct, tokens_in, tokens_out, effort] in config
2. Restart the gateway and send any message
3. Footer renders like: deepseek-v4-flash · ctx 12% · in 15.9k · out 1.2k · effort:high
4. On Feishu it appears as a card-style quote block; other platforms keep the inline format

Checklist

Code

• I have read the Contributing Guide
• My commit messages follow Conventional Commits
• I searched for existing PRs — see Related Issue section
• My PR contains only changes related to this feature
• I have run pytest tests/ -q and all tests pass — full suite blocked by local env (Python 3.9 vs 3.11 requirement); footer logic verified directly via unit invocation
• I have added tests for my changes — verified via direct unit calls
• I have tested on my platform: macOS

Documentation & Housekeeping

• I have updated docstrings in gateway/runtime_footer.py
• Updated cli-config.yaml.example — N/A (no new config keys; reuses existing runtime_footer fields)
• Updated CONTRIBUTING.md/AGENTS.md — N/A
• Considered cross-platform impact — pure Python, no platform-specific code

[Enough1122]

AI code review — automated review for reference, author can ignore or act on any point.

feat(gateway): add opt-in tokens_in/tokens_out/effort runtime footer fields

1. gateway/run.py — effort=(_load_gateway_config().get("agent") or {}).get("reasoning_effort") is evaluated eagerly on every turn, including when the runtime footer is disabled, because it is an argument to build_footer_line(...) which is called unconditionally inside the try. That is a full config YAML re-parse per message just to compute a usually-unused argument. Consider reading the config only when the footer feature is enabled, or caching the resolved reasoning_effort value.
2. gateway/runtime_footer.py — the context_pct field now renders as ctx 42% instead of 42%. This changes the output of an existing, already-enabled field — not just the new opt-in fields — so users who configured context_pct get a different footer after upgrade. If the prefix is intended, a release-note mention would help; otherwise keep the old rendering for context_pct and add prefixes only for the new fields.
3. gateway/run.py — the Feishu-specific card formatting (---\n> {_footer_line}) is hardcoded in the core gateway loop. Platform-specific rendering usually lives in the platform adapter; consider passing the plain footer line and letting the Feishu adapter wrap it, so other card-style platforms can reuse the pattern without editing core.