[TASK] Improve resilience of the `map_zip_with`

[mythrocks]

This is a(nother) follow-up task for #3490, which introduced support for the map_zip_with operation.

When reviewing #3490, one found that for very large maps, some of the data-structures allocated for in the implementation might be too large to fit within the size_type size.

As the algorithm currently stands, it fails when an overflow is detected, to avoid accidental memory corruption.

It would be good to address this, perhaps with batching. This should allow the consumption of m (<n) rows of the inputs at a time, processing them to produce intermediates, and then combining those to produce a final output.