🤖 CoderWorkspaceClaim: a bound claim whose member failed stays Reconciling, so GitOps waits only time out #454

Problem

A bound CoderWorkspaceClaim whose member fails stays Reconciling=True forever, so the GitOps health rules from #453 show it Progressing (Argo CD) or InProgress (Flux) and never Degraded/Failed. A sync wave or a Flux wait: true that waits on such a claim ends only at its timeout.

Reachable member failures after a bind:

• OwnershipUnknown: someone renamed the kwp- workspace, or it has another owner or template (readPoolMemberWorkspace, internal/controller/coderworkspacepoolmember_ready.go), on every poll of a bound member.
• ChangedExternally: someone started a build before the bind extension finished (poolMemberStartBuildStep, reached from watchReady while status.bindExtended is false).

writeClaimStatuses (internal/controller/coderworkspacepool_claim.go) copies the member's reason, but sets Stalled only for Lost or a Share stall. The pool deletes only unclaimed failed members, so the claim stays Bound, not ready, and Reconciling=True.

Options

1. Set Stalled=True on a bound claim whose member phase is Failed (the claim is not Lost, because the workspace can still exist). The health rules need no change.
2. Keep the behavior, which 🤖 feat: GitOps health rules and troubleshooting for workspace pools (#153 PR 16a) #453 documents in docs/how-to/gitops.md: delete the claim and let GitOps create it again.

Found by the independent readiness review of #453 (plan PR 16a of #153).

Generated with xum • Model: anthropic:claude-opus-5-5 • Thinking: xhigh