feat(dashboard): session workflow dashboard MVP (POS-92)

[positivef]

Summary

• Read-only, zero-dependency tool that aggregates multi-session data for one project and renders a self-contained static HTML dashboard (no server, no DB, stdlib only).
• Sources (read-only): .claude/coordination/claude-*.md (slots), ~/.claude/projects/<slug>/*.jsonl (sessions), and git (branch/uncommitted/unpushed/commits-24h/worktrees).
• Detects 5 multi-session "tangles": slot_collision (red), branch_divergence (warn), duplicate_work (warn), stale_slot (yellow), uncommitted_drift (yellow).
• Safety: read-only, idempotent, fail-safe per source, output isolated to a gitignored folder. Git output decoded as UTF-8 to avoid Windows cp949 crashes on Korean commit messages.
• Implements Phase 1 (single-project MVP) of the approved design (Linear POS-92).

Design departures (intentional, documented in README)

• Slots and sessions are separate panels (coordination claude-N and jsonl sessionId share no key).
• duplicate_work uses "2+ in-progress slots" heuristic (per-session files_touched deferred to Phase 2).
• Model embedded in HTML (not fetched) so it opens via file://.

Test Plan

• pytest tests/test_workflow_*.py -> 30 passed
• ruff check workflow_dashboard/ -> clean
• Real smoke run against fieldsync-pro (6 slots, 5 sessions, conflicts detected)
• Rendered HTML verified in a real browser (Korean + emoji render, live session highlighted, no JS errors)

Refs: Linear POS-92

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: c30b89fa3e

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[chatgpt-codex-connector]

Normalize all non-alphanumeric chars in session slug

slug_for() only replaces :, \, /, _, and ., but Claude session directories are encoded from cwd by replacing every non-alphanumeric character with -. For project paths containing other characters (for example spaces or parentheses), the computed slug does not match the on-disk folder, so resolve_jsonl_dir() misses transcripts and the dashboard can incorrectly show zero sessions/conflicts.

Useful? React with 👍 / 👎.