fix: catch exception if unmarshall the rpc request encounters error

[levy5307]

What problem does this PR solve?

In previous implementation, the cluster will crash if we receive a dirty muation log, and will not recover unless the dirty logs are clear. For more details: #617
So in pr(XiaoMi/rdsn#863), I changed the way from coredump to throw exception, and we can catch the exception to avoid cluster crash.

Checklist

Tests

• Manual test (add detailed scripts or steps below)

1. send a dirty request by java client
2. to see whether the perf counter increase
   [replica*app.pegasus*recent_corrupt_write_count@2.4, VOLATILE_NUMBER, 1.00]
3. to see whether there is a log about this situation.
   pegasus_server_write.cpp:106:on_batched_writes(): pegasus batch writes handler failed, from ={ip}:{port}, exception =no more data to read after end-of-buffer