feat: Request all arrays from uproot at once inside dask task

[nsmith-]

So we can do more IO optimizations inside uproot's source (e.g. coalesced reads, more concurrent decompression)

FYI @jpivarski while working on this I found that uproot 5.3.3 broke test_nanoevents_delphes.py and test_nanoevents_treemaker.py.
scikit-hep/uproot5@v5.3.2...v5.3.3 doesn't suggest anything obvious

[nsmith-]

In the longer term we should really refactor nanoevents/mapping. It is a bit convoluted mostly due to being designed to work with https://github.com/CoffeaTeam/columnservice with globally unique form keys. I think we will find a different implementation of column caching. Then, more of the form key logic can be left to uproot (as it already is implemented for the sake of uproot.dask) and we can simplify the mapping to just implement the DSL for things like counts2offsets, parents, etc.

[jpivarski]

@jpivarski while working on this I found that uproot 5.3.3 broke test_nanoevents_delphes.py and test_nanoevents_treemaker.py.
scikit-hep/uproot5@v5.3.2...v5.3.3 doesn't suggest anything obvious

Does Delphes have any std::bitvector TBranches? Before, they would have been skipped; now we'd be trying to interpret them. Maybe that's going wrong?

[lgray]

closes #1077