Add bounded GitHub API timeouts, retries, and sanitized structured errors

[cervantesh]

Maturity context

• Project maturity: operational (66/100)
• Dimension: performance_reliability
• Priority: high

Evidence

• src/api.ts createClient calls fetch without AbortSignal or timeout
• Non-2xx handling embeds the full response text in Error.message
• Response request IDs and rate-limit headers are discarded
• docs/OPERATIONS.md asks operators to record GitHub request IDs when available

Scope

• Add a configurable internal request timeout with a safe documented default.
• Retry only idempotent/read operations and explicitly safe transient failures with bounded exponential backoff and jitter.
• Parse status, request ID, rate-limit headers, and sanitized GitHub error fields into a typed error.
• Redact authorization material and cap response-body text in surfaced errors.
• Expose enough structured detail for recovery without logging automatically.

Acceptance criteria

• A hanging mocked request aborts within the configured timeout.
• 429 and retryable 5xx fixtures follow the bounded retry policy; unsafe mutations are not duplicated.
• Errors include status and request ID when present but never tokens or full sensitive bodies.
• Rate-limit exhaustion produces an actionable reset/retry signal.
• docs/OPERATIONS.md documents timeout, retry, and error fields.

Help wanted

Requires careful reliability and security review; mocked response fixtures make it independently testable.

---

Generated from the repository maturity assessment dated 2026-07-11. Do not include credentials, customer data, or local-only files in public discussion.