fix(worker): eliminate branch UDA — session name now derived from description

[birdmanmandbir]

Summary

• Root cause: SessionName() used task.Branch (preferred) or task.Description (fallback). The branch UDA was written after spawn, so lookup time computed a different session name than spawn time → "no worker session assigned" error.
• Fix: Remove the branch UDA entirely. SessionName() now always derives the slug from task.Description, which is immutable and available at both spawn and lookup time.

Changes

• internal/taskwarrior/taskwarrior.go — remove Branch field from Task struct, remove UpdateWorkerMetadata, fix SessionName() to always use description
• internal/taskwarrior/queries.go — replace branch.any: with spawner.any: in GetActiveWorkerTasks
• internal/worker/spawn.go — remove UpdateWorkerMetadata call (no longer writes branch UDA at spawn)
• cmd/pr.go, internal/pr/pr.go, internal/review/review.go — remove task.Branch fallbacks, use WorktreeBranch() exclusively
• cmd/pr.go merge — fix cleanup guard that was gated on ctx.Task.Branch != "" (would have silently skipped cleanup after this change)
• internal/worker/list.go — use WorktreeBranch() for live branch display; fall back to GenerateBranch() if worktree gone
• internal/tui/task_detail.go — derive display branch from description (display-only, no I/O needed)
• internal/tui/actions.go — remove branch-based fallback path in resolveWorkDir
• internal/worker/hook.go — remove unused hookTask.Branch() accessor

Test plan

• make ci passes
• ttal task advance <uuid> spawns worker — verify no branch: UDA written (task <uuid> export | jq .branch returns null)
• ttal open session <uuid> attaches successfully after spawn
• ttal worker list shows correct session names
• ttal pr merge triggers cleanup (no longer blocked by missing branch field)

[birdmanmandbir]

PR Review — fix(worker): eliminate branch UDA

Summary

This PR fixes the "no worker session assigned" error by removing the race condition between session creation and the branch UDA write. SessionName() now derives deterministically from the task description (immutable), so open/attach commands always find the session regardless of timing.

The fix is sound and the implementation is consistent across all call sites.

---

Critical Issues

None.

Important Issues

None.

Suggestions

[DRY, low-priority] internal/pr/pr.go:14 and cmd/pr.go:49 contain identical two-step branch resolution logic (try WorktreeBranch, zero on error, error if empty). This duplication was previously masked by the Branch UDA fallback. Consider extracting a RequiredBranch(ctx) helper to consolidate it — though the different error messages at each site are intentional, so it's not urgent.

[Clarity] internal/worker/list.go:149 is now the only site that uses a two-tier WorktreeBranch → GenerateBranch soft fallback for display. Worth a comment clarifying it's display-only, to prevent future callers from copying this pattern into non-display paths:

// display only — falls back to generated name when no live worktree exists

---

Strengths

• Root cause is correctly identified and cleanly eliminated — SessionName() is now a pure, stable function
• spawner.any: is a better discriminator for active worker tasks than branch.any: was
• UpdateWorkerMetadata, hookTask.Branch(), and branch-based resolveWorkDir all removed without backward-compat shims (good YAGNI hygiene)
• Task struct is now a clean data record with no derived/mutable state
• Tests updated proportionally to match removed behavior

VERDICT: LGTM

[birdmanmandbir]

Triage Update

Fixed

• Clarity comment on two-tier branch fallback in internal/worker/list.go:149 — added // display only — falls back to generated name when no live worktree exists (commit 29880e3)

Deferred

• DRY: extract RequiredBranch(ctx) helper for internal/pr/pr.go:14 + cmd/pr.go:49 — valid observation, but the two sites have intentionally different error messages making extraction awkward. Low priority, not a correctness issue.