Show agent installation readiness with explicit unknown and retry states

[alimobrem]

The readiness page now presents the agent's structured installation diagnostic separately from the cluster checklist. Operators can see provider configuration/connectivity, database, Kubernetes read permissions, and monitoring observations, with remediation, installation-credential scope, original check timestamps, and a read-only retry.

Unknown provider connectivity and missing or malformed checks cannot become a healthy installation badge. Authentication/unavailable endpoint failures show readiness as unknown; failed refreshes explicitly mark retained observations as a previous report. The paired backend endpoint performs no model request or workload mutation and caches observations for 60 seconds; these checks do not establish model execution or incident recovery.

Validation:

• Full Vitest suite: 196 files, 2,353 tests passed.
• Isolated release acceptance lane: 74 tests passed, including 25 new readiness cases; rerun after rebase.
• TypeScript, new-file lint, production build, and diff checks passed. Existing onboarding files retain five lint warnings.
• No local Pulse stack or real-cluster acceptance was run.