[improve][java-client]Shrink BatchMessageContainer maxBatchSize

[AnonHxy]

Motivation

There will be memory waste when producer publishes messages with batch enable, which may cause some problems, e.g., issue #17662 and issue #14943

The PR #15033 try to fixed the memory waste problem but bringing a significant performance regression.

Here we introduce another way to solve the memory waste problem through shrinking the maxBatchSize of BatchMessageContainerImpl, without bringing performance regression, which will solve the two issues list above.

Following is the test result to approve it：

1. Solved memory waste

Reducing the memory waste will be very helpful for this issue: #17662

After applying this PR, following the steps of the issue to reproduce it, we can see that memory will not exceed the client memory limit:

2. Without bring performance regression

• Here is the publish throughput(msg/s) result, And we can come to conclusion that shrink will not bring performance regression comparing with compositeBuffer

	buffer	shrink	compositeBuffer
1	1,057,249.52	1,038,105.78	917,114.21
2	1,018,946.85	1,050,870.61	896,039.70
3	1,037,839.98	1,038,394.16	903,520.30
avg	1,038,012.11	1,042,456.85	905,558.07

buffer : means using allocator.buffer() to allocate batchedMessageMetadataAndPayload without shrink.
shrink : means applied this PR
compositeBuffer : means applied #15033, which using allocator.compositeBuffer() to allocate batchedMessageMetadataAndPayload

Modifications

• Shrink AbstractBatchMessageContainer#maxBatchSize if meet the conditions.
• Revert allocator.compositeBuffer() to allocator.buffer(Math.min(maxBatchSize, ClientCnx.getMaxMessageSize()))

Verifying this change

• Make sure that the change passes the CI checks.
• Add test org.apache.pulsar.client.impl.BatchMessageContainerImplTest#testUpdateMaxBatchSize

Documentation

• doc-not-needed
  (Please explain why)

PR in forked repository: AnonHxy#5

[tisonkun]

Thanks for submitting this patch! Comments inline.

[AnonHxy]

@tjiuming @Jason918 @merlimat @codelipenghui PTAL thanks

[Jason918]

nice work

[tisonkun]

LGTM. Thank you!

[tisonkun]

/pulsarbot run-failure-checks