Speed up `create_batch_from_map`

[Dandandan]

Is your feature request related to a problem or challenge? Please describe what you are trying to do.

Currently .to_array() is called on each scalar value that creates a single row array, which is very inefficient:

• There is overhead for generating the array structure for each row.
• The single-row arrays are concatenated afterwards at the end, which is slow and would be unnecessary if they are created with more values at once
• Intermediate Vecs are generated, causing more memory usage / allocations / fragmentation.

I expect this should speed up some db-benchmark queries (group by queries with smaller groups) considerably and may decrease memory usage by quite a bit.

Describe the solution you'd like
Iterate over the values and emit arrays of batch_size elements at once.
Or as a first step just do it for all of the values (as is the case currently) - and emit smaller batches in a later PR.

To do it with batch_size there should be some state and/or the groups should be removed from the map.

Describe alternatives you've considered
n/a

Additional context
n/a