Do not cache numba decorated split_array

[dachengx]

What is the problem / what does the code in this PR do

As title suggested. If still using cache=True, you will see errors like

Exception in thread load:event_se_density:
Traceback (most recent call last):
  File "/opt/XENONnT/anaconda/envs/XENONnT_2024.03.1/lib/python3.9/threading.py", line 980, in _bootstrap_inner
    self.run()
  File "/opt/XENONnT/anaconda/envs/XENONnT_2024.03.1/lib/python3.9/threading.py", line 917, in run
    self._target(*self._args, **self._kwargs)
  File "/home/xudc/strax/strax/mailbox.py", line 290, in _send_from
    self.close()
  File "/home/xudc/strax/strax/mailbox.py", line 354, in close
    self.send(StopIteration)
  File "/home/xudc/strax/strax/mailbox.py", line 307, in send
    raise MailboxKilled(self.killed_because)
strax.mailbox.MailboxKilled: (<class 'ValueError'>, ValueError('no field of name cut_s1_single_scatter'), <traceback object at 0x7fee96cc5240>)
    inputs_merged = {
  File "/home/xudc/strax/strax/plugins/plugin.py", line 497, in <dictcomp>
    kind: strax.Chunk.merge([inputs[d] for d in deps_of_kind])
  File "/home/xudc/strax/strax/chunk.py", line 229, in merge
    data = strax.merge_arrs(
  File "/home/xudc/strax/strax/utils.py", line 215, in merge_arrs
    result[fn] = arr[fn]
ValueError: no field of name cut_s1_single_scatter

when multiple jobs run in parallel. The reason for this is deep. But the error log suggest that there is problem of numba function, so that split_array returns wrong results.

Can you briefly describe how it works?

Compile split_array inside each job, do not refer to the cached numba function.

Can you give a minimal working example (or illustrate with a figure)?

Please include the following if applicable:

• Update the docstring(s)
• Update the documentation
• Tests to check the (new) code is working as desired.
• Does it solve one of the open issues on github?

Please make sure that all automated tests have passed before asking for a review (you can save the PR as a draft otherwise).

[coveralls]

coverage: 90.507%. remained the same
when pulling f6a1c74 on no_cache_numba_split_array
into 2d772f1 on master.