[query/vds] Actually use `ref_block_max_length` in `to_dense_mt`

[chrisvittal]

Right now, we perform a full scan in to_dense_mt, we have information to do less work and densify in a single pass.

• Expose partitioning in python
• For each partition in the variants table, use ref_block_max_length to determine the full reference interval necessary to densify that partition
• Use map_partitions of the variants and query_table on the reference to get two streams with all information necessary to densify.
• Join the streams and use the current algorithm/scan to do the work.