[FLINK-38429] Reuse OperationSerializer in Elasticsearch 8 connector to improve perf

[beliefer]

This PR proposes to improve perf for Elasticsearch 8 connector.
Currently, Elasticsearch 8 connector create a new OperationSerializer for each Operation.
A large amount of OperationSerializer's instances brings big overhead.

There is a benchmark cases.

1. Flink application with single parallelism.
   The Kafka Topic consumption rate show below.

Before this PR	After this PR
1.5 K strips/s(min)	2.3 K strips/s(min)
2.0 K strips/s(max)	2.9 K strips/s(max)

2. Flink application with 32 parallelisms.
   The Kafka Topic consumption rate show below.

Before this PR	After this PR
58.3 K strips/s(min)	70.3 K strips/s(min)
60.0 K strips/s(max)	74.5 K strips/s(max)

[beliefer]

ping @mtfelisb @liuml07 cc @reswqa

[liuml07]

The PR makes sense to me. Thanks for fixing this.

I think we can make the commit / PR title more descriptive. How about something like Reuse OperationSerializer in Elasticsearch 8 connector to improve perf