Resiliently handle `karapace` catalog unreachable

[jfallows]

Describe the desired outcome from the user's perspective
As devops, I want zilla to be resilient when karapace catalog is unreachable.

Acceptance criteria

• resilient retries when catalog unreachable
• handle catalog not reachable yet on first access
  • models treat missing schema using reject-messages-by-default strategy
  • bindings treat missing schema using reject-streams-by-default strategy; openapi, asyncapi, grpc
• handle catalog not reachable after already caching a result
  • reuse cached result beyond cache expiration
  • log events indicating behavioral workarounds when unreachable, also when later reachable
  • unify caching across workers to maximize cache hits

Additional context
None