perf: use fast uproot.num_entries method instead of TTree.num_entries

[pfackeldey]

uproot.num_entries is a fast path to get the number of events. It should always be faster than TTree.num_entries. This may become noticeable for preprocessing very large filesets.

The nanoevents factory should use uproot.num_entries aswell (at some point), in this PR I just removed the overhead of calculating it twice.

[ikrommyd]

@pfackeldey I was wondering because I find preprocessing slow (many minutes for large datasets), did you try it and have seen speed improvement?

[pfackeldey]

@ikrommyd I have not measured it, because I don't have an analysis setup to do so. If you can, I'd be happy to see the difference!
I'm not sure how much one can gain here (I didn't profile the preprocess method), but calculating the number of entries should be always faster with uproot.num_entries compared than TTree.num_entries.
(I noticed slow preprocessing aswell when running the AGC - actually I might be able to measure the performance gain with the 200GBps challenge?)