fix: make volume copy cleanup resilient and allow live destinations - #58

Summary

Fix workspace volume-copy cleanup so completed copies do not leave source/destination permanently lifecycle-locked, and allow an operator to copy into a destination workspace that is already running.

Stale Copying fix

Production showed a completed operation (status=succeeded) with two stale workspace_volume_copy_locks. The Kubernetes Job/Pod had already disappeared, while coderd repeatedly received HTTP 403 trying to DELETE jobs.batch and never released the locks.

This PR adds a second, code-level safety net in addition to the RBAC fix in biptec/coder-infra#32:

• DeleteCopyJob checks Job existence before issuing DELETE
• if GET confirms the Job is already gone (for example TTL cleanup), cleanup succeeds without attempting DELETE and locks can be released
• if the Job exists, coderd still performs foreground DELETE and waits until Kubernetes confirms it is gone before releasing locks
• DELETE 404 remains successful, covering the GET->DELETE TTL race
• DELETE 403 while the Job still exists remains an error, so locks are never released prematurely

Running destination

• destination status running is now accepted in addition to stopped
• transitional/error destination states remain blocked
• the UI shows a prominent warning that the destination is running and files may be modified during the copy
• the warning also explains that Start/Stop/Delete and other lifecycle changes remain locked only until the copy completes
• a running destination itself is not stopped by the copy flow

Verification

Backend:

• go test ./workspacevolumecopy ./coderd/workspacevolumecopy ./cmd/workspace-volume-copy-helper — passed (same volume-copy block as custom-release CI)
• go test ./coderd -run '^TestWorkspaceVolumeCopy' -count=1 — passed
• running-destination integration test verifies destination stays Running, Stop is 409 during copy, copy succeeds, then Stop succeeds after cleanup
• cleanup regression tests cover:
  • Job already gone -> no DELETE attempted
  • Job disappears between GET and DELETE -> DELETE 404 succeeds
  • Job exists and DELETE is forbidden -> error retained, no premature unlock

Frontend:

• Biome format/lint on changed files — passed
• Vitest unit tests — 12/12 passed
• TypeScript lint:types — passed
• a Storybook interaction regression was added for the running-destination warning + enabled Copy button; the local browser runner in this dev workspace is killed during Vite dependency bundling before assertions (and reports the repository's existing vitest/@vitest-browser version mismatch), so this is not counted as a passing local browser test

Other:

• gofmt and git diff --check — passed
• branch is based on the current origin/custom/v2.35.3

Dependency for immediate Kubernetes Job deletion: https://github.com/biptec/coder-infra/pull/32

No deployment or merge is included in this PR.