ref(streaming): Refine timeout stream behavior

[iambriccardo]

This PR refines the TimeoutStream behavior to correctly exhibit the intended timeout semantics. Previously, the timer only started when the inner stream returned Pending. Now the timer starts immediately when reset, giving us a fixed window per batch.
The semantics are:

• Timer starts as soon as a new batch begins.
• If the batch isn't filled within max_batch_fill_duration and the stream goes idle, we emit a timeout.
• If data is actively flowing, we prioritize delivery over the timeout, even if the timer has technically expired, we keep returning values as long as they're available.

This prevents the timeout from interrupting a healthy stream that happens to be slow, while still guaranteeing we flush partial batches when the source truly goes quiet.

[coveralls]

Pull Request Test Coverage Report for Build 19759445592

Details

• 12 of 12 (100.0%) changed or added relevant lines in 1 file are covered.
• 24 unchanged lines in 2 files lost coverage.
• Overall coverage decreased (-0.02%) to 82.338%

---

Files with Coverage Reduction	New Missed Lines	%
etl-postgres/src/tokio/test_utils.rs	4	86.47%
etl/src/replication/client.rs	20	85.55%

Totals
Change from base Build 19701980580:	-0.02%
Covered Lines:	16428
Relevant Lines:	19952

---

💛 - Coveralls