feat(chat): add ask_user_question interaction flow

[Pandakingxbc]

Summary

• Add a built-in ask_user_question structured tool for every agent, backed by LangGraph interrupt/resume.
• Persist pending questions in SQLite/PostgreSQL so Dashboard refreshes and process restarts can recover the interaction.
• Support Channel text interaction through /answer; a normal message is accepted automatically only for one unambiguous free-text question.
• Add a Dashboard question card with single choice, multi-select, custom input, skip, question paging, recommended badges, and explicit final submission.
• Reuse the existing chat stream for resume output, nested questions, nested tool approvals, usage tracking, and completed-turn bookkeeping.
• Keep ask_user_question outside tool-approval HITL rules to avoid approval deadlocks.
• Add English and Chinese tool, slash-command, card, and error copy.

Product behavior

• One to three questions are supported per tool call.
• A single-choice click advances to the next question; the final question still requires Submit.
• Multi-select questions require Continue or Submit.
• Channel choice questions use /answer <number-or-label>; batches use /answer 1=value; 2=value.
• Pending records use an atomic pending -> resuming -> answered transition, with release on resume failure and recovery after restart.

Validation

• uv run ruff check src tests
• uv run mypy src/octop
• Relevant backend suites: 115 passed, 1 skipped; added Channel projection suite: 5 passed
• Dashboard focused tests: 6 passed
• npx eslint on all changed Dashboard TypeScript files
• npx tsc --noEmit
• npm run build
• Dashboard full suite: 449 passed, 1 unrelated existing failure in publishedExperts.test.ts because its expectation omits the refresh request body already sent by main

The repository-wide backend suite was stopped at 12% as requested; no failure had appeared before interruption.