improve zap chunking

[mschoch]

chunk mode 1025 made singinifcant improvement for indexes
with many unique fields (or very low cardinality)

here we introduce chunk mode 1026 which takes the same
approach as 1025, but extends it to improve chunking for
all other cardinalities

specifically, we try to create the fewest number of dense chunks
chunk size is capped at 1024 documents as before

[mschoch]

For some reason this produces larger indexes.

[sreekanth-cb]

Tried indexing wikipedia text for about 2.3M documents with the default mapping.
Size in v13 format - 12.69GB single segment index.

Size with v14 format - 12.26GB single segment index.
So ~430MB size reduction observed.

Also size reduction noted between a multi segment index to single segment index here - 12.79 => 12.26 GB with v14.

@mschoch , any context about the size increase you observed?

[mschoch]

@sreekanth-cb using bleve-blast to index 1 million documents I got the following:

$ du -sh bleve-scorch-zapv14-mil.bleve 
5.9G	bleve-scorch-zapv14-mil.bleve
$ du -sh bleve-scorch-zapv13-mil.bleve 
3.5G	bleve-scorch-zapv13-mil.bleve

It's not single segment, so probably a merge was in progress. I have to update bleve-blast to have an option to use the IndexBuilder.

[mschoch]

@sreekanth-cb nevermind, my bleve-blast silently revereted to zapv11 again because we still have not merged this fix:

blevesearch/bleve#1401

[mschoch]

Confirmed small improvement with v14 with IndexBuilder on wiki dataset and confirmed the integration tests pass with v14, so will proceed to merge this, and do the bleve/zap release dance.