feat(feishu): add native update prompt cards

[kshitijk4poor]

Salvage of #21911 by @heathley onto current main.

What this does

Adds send_update_prompt() to the Feishu adapter so when hermes update --gateway needs y/n confirmation (stash restore, config migration), Feishu users get an interactive Yes / No card with proper authorization gating instead of the plain-text fallback.

Today, gateway/run.py does getattr(type(adapter), "send_update_prompt", None) is not None and falls back to plain text when the method isn't defined. Discord, Telegram, and qqbot (#21353) already implement it — Feishu was the gap.

Implementation

Mirrors the existing Feishu approval-card architecture (send_exec_approval / _handle_approval_card_action / _resolve_approval / _build_resolved_approval_card):

• _update_prompt_state dict + counter for prompt-id tracking
• send_update_prompt() builds an interactive Card V2 payload, sends via _feishu_send_with_retry, stores state on success
• _handle_update_prompt_card_action() handles button clicks: validates prompt_id, validates answer (y/n), authorizes operator against existing _admins and _allowed_group_users allowlists, schedules resolution, returns a resolved-state replacement card
• _resolve_update_prompt() pops state and atomically writes ~/.hermes/.update_response (matches qqbot's _write_update_response semantics)
• Method signature (chat_id, prompt, default, session_key, metadata) matches Discord's exactly

Bonus: _submit_on_loop now returns bool so callers can detect scheduling failures (loop closed, RuntimeError) and avoid showing a card as resolved when the resolution coroutine couldn't actually be scheduled. The existing approval handler picks up the same fix.

Tests

13 new test cases integrated into tests/gateway/test_feishu_approval_buttons.py:

• Send: success, state stored, not-connected, send failure
• Callback: yes, no, missing prompt_id, already-resolved, schedule failure, unauthorized operator
• Resolve: writes response file, overwrites existing file, drops unknown prompt_id silently

Full Feishu test suite: 256 passed across test_feishu_approval_buttons.py, test_feishu.py, test_feishu_onboard.py.

Why salvage

PR #21911 was 486 commits behind main, but only 3 unrelated commits had touched feishu.py since (monotonic deadlines, topic-thread cleanup, topic-reply threading) — none overlapped with the PR's hunks. Cherry-pick auto-merged cleanly with +433/-3 matching the PR's stats exactly. @heathley's authorship preserved.

Closes #21911.

Use --rebase merge to preserve @heathley's authorship.

[github-actions]

🔎 Lint report: salvage/feishu-update-prompt-21911 vs origin/main

ruff

Total: 0 on HEAD, 0 on base (➖ 0)

🆕 New issues: none

✅ Fixed issues: none

Unchanged: 0 pre-existing issues carried over.

ty (type checker)

Total: 7876 on HEAD, 7876 on base (➖ 0)

🆕 New issues: none

✅ Fixed issues: none

Unchanged: 4171 pre-existing issues carried over.

Diagnostics are surfaced as warnings — this check never fails the build.