test: stabilize connect e2e (#3391)

## Summary
- Stabilize two Connect E2E cases from nightly run `26712668301`.

## Why
- `test_exactly_once_source` used a 1s `read_committed` validation
consumer timeout. In the failed run, the first FETCH returned committed
records a few milliseconds after the consumer timed out.
- `test_pause_and_resume_sink` paused the sink while the task was
RUNNING but before the sink consumer had completed
assignment/reconciliation and consumed any records.

## What Changed
- Increase the Connect exactly-once validation consumer timeout from 1s
to 5s while keeping `read_committed`.
- Wait for the Connect verifiable sink to consume at least one record
before pause/resume.

## Verification
- `python3 -m py_compile
tests/kafkatest/tests/connect/connect_distributed_test.py`
- `git diff --check`
- Local ducktape:
- `ConnectDistributedTest.test_exactly_once_source` with `clean=false,
connect_protocol=eager, metadata_quorum=ISOLATED_KRAFT,
use_new_coordinator=true`: PASS, session `2026-06-01--003`, runtime
`6m21s`.
- `ConnectDistributedTest.test_pause_and_resume_sink` with
`connect_protocol=compatible, metadata_quorum=ISOLATED_KRAFT,
use_new_coordinator=true, group_protocol=consumer`: PASS, session
`2026-06-01--002`, runtime `1m50s`.

codex autobalancer decision trace query

feat(release): release 1.6.6

fix(release): bump kafka version to 3.9.1 (#3296)

Signed-off-by: Robin Han <hanxvdovehx@gmail.com>

feat(release): release 1.6.4-rc1 (#3195)

Signed-off-by: Robin Han <hanxvdovehx@gmail.com>

feat(release): bump version to 1.6.4-rc0 (#3180)

Signed-off-by: Robin Han <hanxvdovehx@gmail.com>

1.6.31.6.31.6.3