* Refactor gitparse for reduced peak memory use
Instead of one single `git log` that grows huge, get a commit list with
"log" then request `git show` groups of those commits.
This is currently slightly slower than main, and parallelizing this part
further doesn't improve things because of the diffChan consumer in the
caller. It does pipeline just enough that one git show can run while
scanning consumes the prior set of diffs.
* Group commits while reading instead of slurp and split
* Make the low-memory scan mode a non-default option
meredith
4b7d1d3a68