fetch: add retry/backoff and endpoint resilience for Overpass requests

[blackopsrepl]

Problem

The default configuration relies on a single Overpass endpoint and a single request attempt.

Why this matters

Live Overpass infrastructure is rate-limited and occasionally unreliable; production consumers need better resilience and diagnosability.

Scope

• Add configurable retry/backoff.
• Support endpoint fallback or endpoint pools.
• Improve error context for HTTP failures and timeouts.
• Consider cache reuse behavior when refresh attempts fail.

Acceptance criteria

• Transient request failures are retried.
• Multiple endpoints can be configured.
• Errors include enough detail to diagnose endpoint/network problems.

[blackopsrepl]

Resolved by #21.

Overpass fetches now support configurable endpoint pools, retry/backoff for transient failures, and richer endpoint-specific error context when all attempts fail.