Fix potential dataloss

[accountas]

Changes

• Check running flag before swallowing interrupt exception in the commit thread. Flag is false after interrupting from the happy shutdown path. If flag is true exception is unexpected and should be propagated.
• Remove the cause check from isInterruptionException. Swallowing interrupt exception this way is a bad practice, so we can expect this to never happen. At the moment it's too aggressive.

Why?

Current implementation can cause data loss. For example, in our environment while loading to Google Cloud Storage we observed the following exception being thrown from time to time:

com.google.cloud.storage.StorageException: Read timed out
...
Caused by: java.net.SocketTimeoutException: Read timed out

SocketTimeoutException is a subclass of InterruptedIOException which is caught by the isInterruptionException.

This causes the commit to fail, the exception being completely ignored (debug log is the only indication) and loading to continue by skipping the batch completely.