Use "batched" version of git rm

[yarikoptic]

ATM, even though not downloading any context, it takes awhile to run some updates. Eg. is taking almost 20h for an update for 001169, which 13578 files changed, 13578 insertions(+) .

I think part of the issue is that we are issuing an individual git -c receive.autogc=0 -c gc.auto=0 rm -f --ignore-unmatch -- which would need to read/modify/save git index which is quite heavy. there is no git rm --stdin but there is git update-index --remove-force --stdin so we could in parallel to it run rm to remove the file and then feed that path to git update-index --remove-force --stdin. I am not yet 100% certain it would provide for the ultimate speed up, as I think groupping of individual operations might provide better performance, but it might be worth to check. We do use --batch mode for many annex commands, but seems so far do not use any --stdin of git.