StackOverflowError when retrying reading from the middle of window with large amount of data

When there is an exception while reading a window, we restart it with TaskState pointing to last correctly read change. After restarting, we read the entire window from beginning, but we ignore unnecessary rows (up to last correctly read change).

The problem with current implementation is that it does this ignoring recursively (see last findNext at line 148):
https://github.com/scylladb/scylla-cdc-java/blob/a2c3c1823d1f5c210bd210e2d048de0f02ceeb93/scylla-cdc-driver3/src/main/java/com/scylladb/cdc/cql/driver3/Driver3WorkerCQL.java#L104-L151

I have observed one instance of this causing StackOverflowError. What's bad is that it won't be able to succeed upon restart (it will try to do it again after restart and fail in the same fashion).

The long term fix for this is to implement smarter resuming of reading (do not skip client-side, but craft such CQL queries that will read the correct portion of window).

/cc @haaawk Would you approve of a quick fix for this problem (wrapping findNext with while to do the skipping iteratively, not recursively) - doing "smarter resuming" will require more time and there are other priorities right now. Of course I will do "smarter resuming" later.

Start in scylla-cdc-driver3/src/main/java/com/scylladb/cdc/cql/driver3/Driver3WorkerCQL.java, examining the findNext logic around lines 104-151 and how retry state skips already-read changes. Confirm that skipping a large number of rows no longer relies on recursive calls and that retrying a window can complete without StackOverflowError.