CASSANALYTICS-116 Fix ByteBuffer flip() in StreamBuffer.copyBytes() causing data corruption

[jyothsnakonisa]

No description provided.

[bbotella]

Looks good to me

[jyothsnakonisa]

VertxStreamBuffer.copyBytes() was calling flip() after each write, resetting buffer position to 0. When multiple chunks were written to the same buffer, subsequent chunks overwrote previous data instead of appending, causing Stream API failures.

Example:

• Chunk 1: Write "AAAA" at position 0 → flip() → position resets to 0
• Chunk 2: Write "BBBB" → overwrites "AAAA"
• Result: Corrupted data

Solution:

1. Remove flip() from StreamBuffer.copyBytes() - This method writes data chunks and should only advance position, not flip. The caller is responsible for flipping after all chunks are written.
2. Ensure flip() in CdcRandomAccessReader.rebuffer() - After writing all data, flip() transitions the buffer from write mode to read mode by resetting position to 0 and setting limit to the valid data length. This ensures RandomAccessReader can correctly read the buffered data.

tests passed without flip() because:

• Buffers were always completely filled (enforced by assert buffer.remaining() == 0)
• RandomAccessReader#reBufferAt resets position to 0 regardless of flip

However, flip() remains essential for correctness in partial-fill scenarios (e.g., last buffer in file, or when less data is available than buffer capacity).

[yifan-c]

We should avoid the changes in the analytics-sidecar-client and the other sidecar-copied subprojects at the best. Those files are supposed to be in sync with Sidecar.
I would revert the changes in this file and the test file. The VertxStreamBuffer and its test in another package.

[yifan-c]

Thanks for adding the docs. Let's actually add the docs in org.apache.cassandra.spark.utils.streaming.StreamBuffer instead, not this file.

[yifan-c]

I believe the description is no longer accurate. Please update to reflect the actual fix.