Support batchsize/process_batch in the minHash stage

Is your feature request related to a problem? Please describe.
GPUs typically do better with larger batches (1-3GB) for processing but CPU stages do better with smaller (256MB or so) batches or run into OOMs.

To avoid this we'd like to implelement/support proccess_batch for the MinHashStage to ensure that we can use a smaller blocksize during read but saturate the GPU better for minhash.

Describe the solution you'd like
Implement process_batch for the minHash stage that reads the files represented by all the incoming task.data one by one and then concat it. There's still some penalty of having to read it one by one (read not saturated), but the minhash and write can be on larger blocks.

Describe alternatives you've considered
Similar approach to #1332 but arguably simpler since it's the typically process/process_batch type stage.

Additional context
Add any other context or screenshots about the feature request here.

Start with the MinHashStage entry point and inspect how its existing process path handles task.data. Compare the process/process_batch pattern described in issue #1332, then define done as process_batch reading each represented file, concatenating the inputs, and allowing larger MinHash and write blocks while retaining smaller read blocks.