fix: first concurrency issue in Python 3.13t

[jpivarski]

This:

executor = ThreadPoolExecutor(max_workers=16)

for array in uproot.iterate(
    [
        "/home/jpivarski/Downloads/Run2012B_DoubleMuParked.root:Events",  # https://opendata.cern.ch/record/12365
        "/home/jpivarski/Downloads/Run2012C_DoubleMuParked.root:Events",  # https://opendata.cern.ch/record/12366
    ],
    decompression_executor=executor,
    interpretation_executor=executor,   #  <--- especially this one
    handler=uproot.MemmapSource,
    step_size="1 GB",
):
    print(array)

was causing this:

  File "/home/jpivarski/storage/data/python-3.13-uproot/uproot/interpretation/numerical.py", line 108, in final_array
    basket_array = basket_arrays[basket_num]
                   ~~~~~~~~~~~~~^^^^^^^^^^^^
KeyError: 290

in Python 3.13 with free-threading. But this PR fixes it.

Amusingly, it introduces a global lock. But it's a global lock on basket_array dicts, not all objects everywhere.

I've been monitoring the performance and this is even better than ignoring the problem (i.e. don't update the output array—output random bytes—in the rare cases in which the key is missing). This PR is faster and preserves correctness.