Update of StreamToDataSet block

[drslebedev]

• Support several parallel DataSet accumulation
• Support sample accumulation between work() iterations
• Update unit tests

Notes:

• n_pre must be less than the size of the output port's CircularBuffer. The blok waits until all n_pre samples can be written to the output in a single iteration.
• In stream-to-stream mode, the "stop" Tag is not included in the output.
• Stream-to-stream mode has limited support for overlapping triggers. Configurations such as Start-Stop-Start-Stop work as expected, but configurations like Start-Start-Stop-Stop result in undefined behavior. If data accumulation is active, the second "start" is ignored, and the output may vary depending on subsequent tags.
• In stream-to-dataset modes, the "stop" Tag is included in the DataSet output.
• Tags for pre-samples are not stored and are not included in the output.
• It is expected that "start" and "stop" Tags will come from different iterations. If multiple "start" Tags arrive in a single merged tag, only one DataSet is created. The same applies for "stop" Tags. This may lead to incorrect or unexpected output.
• Currently used Matcher doesn't really support overlapping DataSets

Example of Stream-to-Dataset with overlapping datasets and pre/post samples:

[RalphSteinhagen]

LGTM! 👍

I like the way you refactored the processBulk(..) into two independent sub-versions, one for 'stream-to-stream' and the other 'stream-to-dataset' this makes the testing much easier. Also, the latter allows for (partially overlapping) start-stop sequences to produce independent DataSets.... well done.

Thanks also for documenting the edge-cases that are presently UB. Perhaps, it would be better to transfer these from PR/code-only documentation to proper Doc<"..."> doc string. This could be done in a follow-up commit thought.