ticdc: kafka-consumer may oom if upstream workload produce large msg

[AkiraXie]

What did you do?

1. open simple protocol changefeed
2. open claim check in storage
3. open kafka-consumer
4. run workload with row-wide=100-800kb

What did you expect to see?

consumer is ok and sync data to target correctly

What did you see instead?

consumer oomed and pyroscope shows that mysql driver gets large memory

Versions of the cluster

v8.5.1

[AkiraXie]

/severity major

[AkiraXie]

/label affects-8.5