[MISC] Search improvements

[h-2]

NOT FOR MERGE

@eseiler some inspiration for performance tuning of the search. This PR improves unidirectional_search_all/highErrorReadsSearch0 by 20% (impact on complex searches is expectedly smaller).

Left to do in this domain:

• also make the vector-of-cursors an out-parameter (in the internal interface)
• add a public search_buffers config element so that all of the buffers can be reused and no dynamic allocation happens in the search at all.

I expect the latter to have an even stronger influence in parallel contexts where allocations are notoriously slow.