TUI: add full-screen interactive TUI for PR triage and review

[potiuk]

Add a rich full-screen terminal UI for the breeze pr auto-triage command
that replaces the sequential review flow when a TTY is available.

TUI Features

• Full-screen PR list with columns: PR#, Classification, CI Checks,
  Behind, Title (with labels), Author, LLM status, Suggested action,
  Performed action
• Detail panel showing PR info, classification, LLM review status
  with timing, labels, assessment summary, violations, unresolved threads
• Inline diff viewer with syntax highlighting and scroll support
• Tab-based focus cycling between PR list, detail, and diff panels
• Mouse support for clicking rows and scrolling panels

PR Classifications

PRs are automatically categorized: WF Approval, CI Issues, LLM Warnings,
LLM Errors, All passed, Wait for LLM, Stale Review, Recently Triaged.
Each has context-sensitive suggested actions derived from the PR state
(e.g. merge conflicts → draft, 1-2 CI failures → rerun, only unresolved
comments → comment).

Keyboard Shortcuts

• Enter: Sequential assessment (triage) / Detailed review (review mode)
• Space: Toggle select | b: Batch action
• Quick actions: o Open, s Skip, a Approve, d Draft, c Comment, z Close,
  f Rerun failed, r Rebase, m Mark ready, x Flag suspicious, l LLM review
• Navigation: j/k Up/Down, n/p Pages, Tab focus, e Diff, Esc/q Quit

Background Operations

• LLM reviews run in background threads with running/queued distinction,
  per-PR timing (took/queued/attempts), retry (3 attempts, 3min timeout),
  on-demand trigger via l key, and auto-submission for passing PRs
• PR status refresh after actions (5s delay then 60s polling for pending CI)
• Diff prefetching for current and nearby PRs
• Lazy pagination loading more PRs when scrolling near end of list
• Cache validation at startup — batch-checks head_sha via GraphQL,
  removes stale entries

Review Mode

Review mode (--mode review) uses the TUI with green color scheme and
adapted UI for code review workflow:

• PR filtering: Only shows PRs with 'ready for maintainer review' label
  and PRs where the reviewer has previously accepted or requested changes
• LLM action: Uses review_pr (detailed code review with line-level
  comments) instead of assess_pr (triage quality check)
• Review-oriented actions: LLM review and Comment are primary quick
  actions; suggested actions show "ready to review", "review findings",
  "needs rebase" etc. instead of triage-oriented "draft", "close"
• Footer labels: "Detailed review" instead of "Sequential assessment",
  "Review action:" instead of "Quick action:"
• Sequential review mode preserved for non-TTY/CI environments

CLI Changes

• --authors replaces --include-collaborators (choice: contributors/collaborators/all)
• --llm-use replaces --check-mode (choice: both/api/llm)
• --llm-concurrency defaults to CPU count (capped at 8), hardcoded 4 for screenshots
• Reorganized help groups: Mode, Select people, Target selection, Filter options,
  Pagination and sorting, Assessment options, Other

Session Summary

On exit, prints a table of acted PRs with links, suggested/performed actions,
per-PR timing, startup/interactive/LLM time, velocity (PRs/hr), and
LLM avg/median duration.

Code Changes

• Refactored pr_commands.py into extracted modules: pr_cache.py, pr_models.py,
  pr_display.py, pr_comments.py, tui_display.py
• AuthorFilter enum for author selection logic
• _PRRefreshManager for background PR status refresh
• _RefreshResult dataclass for refresh results
• _review_pr_as_assessment() wrapper to use review_pr results in the
  existing LLM pipeline (wraps dict result as PRAssessment)
• Mode-aware get_suggested_action() and get_available_actions() with
  separate triage/review implementations
• Reduced GitHub API timeouts (10s connect, 20-30s read) with error handling
• Skip LLM for bot PRs (dependabot, renovate, github-actions)
• Model discovery via Claude CLI config when API key not set
• Graceful shutdown: cancel futures, 10s countdown, force exit
• Comprehensive documentation update in 13_pr_tasks.rst

---

Was generative AI tooling used to co-author this PR?

• Yes — Claude Code (claude-opus-4-6)

Generated-by: Claude Code (claude-opus-4-6) following the guidelines

[choo121600]

Wow😮
Cool!

[potiuk]

@choo121600 -> it's based on #63770 - I should merge it first :)

[andreahlert]

Security concern with credential handling in _get_claude_cli_api_key.

[github-actions]

Backport failed to create: v3-1-test. View the failure log Run details

Note: As of Merging PRs targeted for Airflow 3.X
the committer who merges the PR is responsible for backporting the PRs that are bug fixes (generally speaking) to the maintenance branches.

In matter of doubt please ask in #release-management Slack channel.

Status	Branch	Result
❌	v3-1-test

You can attempt to backport this manually by running:

cherry_picker 48f0a72 v3-1-test

This should apply the commit to the v3-1-test branch and leave the commit in conflict state marking
the files that need manual conflict resolution.

After you have resolved the conflicts, you can continue the backport process by running:

cherry_picker --continue

If you don't have cherry-picker installed, see the installation guide.