fix: preserve replay budget pauses across restarts (Development)

[CountableNewt]

Development Wire's exhausted replay incident opens another archive request after each restart, then rejects the response body. A paused restart also fails to restore saved retry/range/ETag evidence. Reject archive requests before opening them once incident capacity is exhausted and restore transport evidence from every retained checkpoint state.

Paused lanes continue to return readiness 503 while liveness and database-based startup remain successful. Update the misleading README and cover lane/controller readiness and lease-waiting standby behavior. No hosted budget, source generation, checkpoint, recovery-state, or health-policy changes.

The recovery runbook records the September 7 stalled checkpoint and distinguishes it from Development viewers' Production Corpus feed. Original download buckets have expired, so the original 5 GiB cannot be apportioned between useful progress and repeat downloads. Recovery remains a separate reviewed action requiring provider coverage and byte estimates; this patch does not close the incident.

Validation:

• Full Go suite with a dedicated, migrated local PostgreSQL database: 134 passed, zero skipped/failed.
• Full Go race suite passed (11 optional PostgreSQL fixtures skipped in that run; covered by the separate PostgreSQL run).
• gofmt and git diff --check passed.
• Read-only hosted snapshot and unauthenticated Dev/Prod generation-age comparison documented in the runbook; no authenticated acceptance claim.

Tracks TSW-126 under TSW-92. Preserve Development-first promotion; this PR must not be merged or deployed automatically.

[linear-code]

TSW-126