Regression: worktree add remains silent beyond five minutes before retry succeeds

[chubes4]

Problem

A current workspace worktree add for one Homeboy branch emitted no stdout or stderr and exceeded a 300-second caller timeout:

studio wp datamachine-code workspace worktree add homeboy fix/13093-rustfmt-main \
  --from=origin/main \
  --task-url=https://github.com/Extra-Chill/homeboy/issues/13093 \
  --reuse-policy=isolated \
  --purpose=manual_agent_task \
  --cleanup-policy=remove_on_success

After timeout, targeted worktree get returned 404 and no directory existed. Retrying the identical command with a larger caller bound succeeded. No lock owner, queue position, active phase, durable request identity, or safe retry timing was emitted.

This regresses the bounded/observable add contracts from #1023 and #1026 on a workspace with roughly 124 worktree-like directories.

Expected

• Persist and print operation identity before lock or capacity waiting.
• Emit bounded phase/lock/queue progress before normal caller timeouts.
• Enforce the aggregate operation deadline and return typed retryable status.
• Timeout/retry is idempotent and states whether any reservation, branch, directory, or metadata committed.
• Same-repository serialization remains observable; independent repositories proceed concurrently.
• Deterministic acceptance coverage holds the relevant lock and proves bounded output plus clean retry.

AI assistance

OpenAI GPT-5.6 Sol via OpenCode executed the timed-out and successful retry, reconciled disk/registry state, deduplicated against #1023 and #1026, and drafted this report. Chris Huber is responsible for the issue.