feat(core): add RetryManager state machine for TAPI backoff/retry

[abueide]

Summary

• Add RetryManager class with three states: READY, RATE_LIMITED, BACKING_OFF
• Handles 429 rate limiting with Retry-After header parsing and configurable max intervals
• Implements exponential backoff with jitter for transient errors (5xx)
• Uses sovran store for state persistence across app restarts with validation on restore
• Supports eager/lazy retry strategies for concurrent batch error consolidation
• Includes autoFlush callback for automatic retry scheduling
• Uses getState(true) (queue-safe) for all state reads to prevent race conditions between concurrent canRetry/handle429/handleTransientError calls
• Consolidated handleError/handleErrorWithBackoff into a single unified method with computeWaitUntilTime function parameter — eliminates duplicated logic
• Side effects (logging, Math.random) extracted from dispatch reducers for purity
• Returns RetryResult type ('rate_limited' | 'backed_off' | 'limit_exceeded') from handle429/handleTransientError so callers can detect when retry limits are exceeded
• transitionToReady clears auto-flush timer to prevent spurious flushes
• isPersistedStateValid validates state string against known values
• Add barrel export and test helper utilities

PR 3 of 5 in the TAPI backoff/retry stack. Depends on #1157. Tests in #1159.

Test plan

• All 416 existing tests pass (no RetryManager-specific tests on this branch — those are in PR test(core): add comprehensive RetryManager test suite #1159)
• No TypeScript compilation errors

🤖 Generated with Claude Code