MINOR: Batch the produce in Connect integration tests

[mingyen066]

EmbeddedKafkaCluster#produce only accepts one record and blocks on
send().get(), so callers that need many records have to loop and pay a
round trip per record. Each of those round trips also waits out
linger.ms for a batch that can never fill, because the caller is blocked
in get(). Six call sites across four integration test classes produce
~11,000 records this way.

Add a produce(List<ProducerRecord<byte[], byte[]>>) overload that sends
the whole list before awaiting any acknowledgement, and have the
single-record overloads delegate to it. Blocking behaviour is unchanged.

Measured with -PmaxParallelForks=1, two runs per side:

before   after
TransformationIntegrationTest#testFilterOnTopicNameWithSinkConnector
34.9s   14.6s
ExampleConnectIntegrationTest#testSinkConnector
29.1s   14.8s
TransformationIntegrationTest#testFilterOnTombstonesWithSinkConnector
24.4s   14.4s
ErrorHandlingIntegrationTest#testErrantRecordReporter
13.5s    7.2s
ErrorHandlingIntegrationTest#testSkipRetryAndDLQWithHeaders
12.2s    7.4s
MonitorableSinkIntegrationTest#testMonitorableSinkConnectorAndTask
9.7s    3.7s

Total for these four classes: 149.2s -> 87.4s.