fix: bind operations to cluster identity and correct agent controls

[alimobrem]

Cluster switching could show cached objects and watches from one cluster while actions targeted another. Bind queries, watches, permissions, dialogs, delayed mutations, YAML, logs and terminals to their cluster; reset retained discovery/Argo/terminal state and reject unknown targets. Deployment-local agent actions are blocked for remote resource context instead of implying remote authority.

Remove the implicit scale-to-zero operation before DELETE, preserve failed bulk-delete rows, and add identity/version preconditions to generic deletions. Repair repeated chat reconnect failures and explicit disconnect handling. Bounded chat requires confirmation when no trustworthy category is supplied. Monitor controls show actual server policy and fetch authoritative pause state.

Audit all 27 tracked Markdown files against source, removing stale toolchain, deployment and security claims; label historical design documents and correct component examples.

Validation: all 2,307 tests across 193 files passed; TypeScript and production build passed; lint passed with zero errors (900 existing warnings). New regressions cover same-name cluster resources, delayed callbacks, late old-watch events, retained terminal/Argo state, retry exhaustion/cancellation and server pause state. Rebase onto merged main preserved the tested file tree. Real multi-cluster acceptance remains pending.