merge and join with a large datasets blocks

[modilabs-starscream]

Merging or joining with over 100k rows and ~50 columns when adding a new row (even if empty) process takes maxes CPU and blocks.

Occurs even without calculations.

(try with NMIS water dataset)