Simplify class docstring to describe current architecture without
referencing SDD deviations. Remove redundant inline comments, compact
JSDoc to single-line where appropriate, and ensure all comments use
present tense.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

…gnaling

- Use getState(true) for queue-safe reads to prevent race conditions
  between concurrent canRetry/handle429/handleTransientError calls
- Consolidate handleError and handleErrorWithBackoff into a single
  method that accepts a computeWaitUntilTime function
- Extract side effects (logging, Math.random) from dispatch reducers
- Return RetryResult ('rate_limited'|'backed_off'|'limit_exceeded')
  from handle429/handleTransientError so callers can drop events on
  limit exceeded
- Clear auto-flush timer in transitionToReady
- Validate state string in isPersistedStateValid

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

Replace string literals with TypeScript enums:
- RetryState enum (READY, RATE_LIMITED, BACKING_OFF)
- RetryResult enum (RATE_LIMITED, BACKED_OFF, LIMIT_EXCEEDED)

Extract helper methods for clarity:
- resolveStatePrecedence(): handles 429 taking priority over backoff
- consolidateWaitTime(): uses switch statement for clear wait time logic
- getStateDisplayName(): maps state to display names

Benefits:
- Type-safe state handling (no magic strings)
- Switch statements make control flow explicit
- Each helper method has a single, named responsibility
- Easier to test and maintain

Co-Authored-By: Claude Sonnet 4.5 <noreply@anthropic.com>

Use Object.values(RetryState).includes() instead of maintaining a
duplicate Set of valid states. More idiomatic TypeScript and eliminates
maintenance burden of keeping Set in sync with enum.

Add 29 tests covering all RetryManager states and transitions:
canRetry, handle429 rate limiting, handleTransientError backoff,
reset, retry strategies (eager/lazy), autoFlush callbacks, and
mixed error scenarios.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

Updated test to verify that 429 now correctly overrides BACKING_OFF state
rather than being silently ignored. The 429's retry-after takes precedence.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

- Add test verifying 429 Retry-After overrides long transient backoff
- Track RetryManager instances in autoFlush tests and destroy in
  afterEach to prevent timer leaks

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

Update log assertions for consolidated limit-exceeded message. Add tests
for: RetryResult return values, jitter calculation, isPersistedStateValid
clock skew detection, handle429(0) edge case, and strengthened assertions
that verify behavioral state after clamping/rejection (not just log output).

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>