fix: code review — thread safety, correctness, and robustness improve…

[leonlee]

…ments

• C1: NoOpConnectionProvider throws UnsupportedOperationException instead of returning null
• C2/C3: align markDeferred to NEW status in JDBC store, matching InMemory store semantics
• M1: ExponentialBackoffRetryPolicy validates baseDelayMs > maxDelayMs
• M2: DefaultInFlightTracker uses System.nanoTime() for monotonic TTL expiry
• M3: Outbox.close() logs each component failure at SEVERE before re-throwing
• M4: expand OutboxDispatcher shutdown comment explaining queue clearing safety
• M5: document MicrometerMetricsExporter close-after-recording ordering requirement
• M6: tighten InFlightTracker eviction to every ~256 acquires at 1x TTL threshold
• M7: DispatchResult.RetryAfter rejects zero delay; update test expectation
• M8: fix maxAttempts Javadoc from >= 1 to >= 0
• m1: add Long type branch in JdbcTemplate.bindParams
• m2: fix stale io.elestyle.outbox.spi.JsonCodec references in Javadoc
• m3/m4: extract TERMINATION_TIMEOUT_SECONDS constant in OutboxPoller
• m5: document MySqlOutboxStore transactional requirement for claimPending
• m6: document DefaultInFlightTracker negative TTL behavior
• m7: add idempotency guard to OutboxDispatcher.close()
• m8: document Outbox.start() thread-safety
• n1-n3: clarify workerCount=0 comment, remove redundant condition, document gauge fields
• P2 (review): switch DefaultInFlightTracker to nanoTime for monotonic expiry
• P3 (review): allow baseDelayMs == maxDelayMs (fixed-delay config)
• P3b (review): revert zero-delay rejection on RetryAfter (consistency with RetryAfterException)
• P2 (review): defaultJsonCodec registers custom JsonCodec beans as global default via ObjectProvider
• P3 (review): fetchPendingRows returns null when capacity disappears to prevent false-zero lag gauge
• P3 (review): add idempotency guard to Outbox.close() for double-close from lifecycle + destroyMethod