Fix hitlet splitting

[WenzDaniel]

What is the problem / what does the code in this PR do
In this PR we are adding a small fix and test for a bug found during processing.

The bug itself manifested itself as

ValueError: Want to add a new zero-length peak after splitting!

while the actual problem was within strax.get_hitlet_data(). So here is what happened, lets start with how the algorithm is suppose to work:

During the building of hitlets we join overlapping hits into hitlets and extend those according to the left/right hit extension. This hit rextension may reach into none-recorded regions outside of our recorded data. Therefore, in the next step get_hitlet_data loops over all records and hitlets and chops all hitlets according to the recorded regions before adding the data information to hitlets.

Here is why things went south:

In chunk 57 of run 029728 we had a hitlet for which the chopping of the none recorded regions did not work (purple hitlet in channel 1. Paired with a 4 sigma downward fluctuation of the baseline lead to a hitlet which was complete outside of any recorded region which led to a zero-length hitlet during splitting (purple dashed part).
The chopping failed because before chopping, get_hitlet_data searchs channel independent for all records which overlap with the corresponding hitlet. In a second step all records which channel does not match the one of the hitlet are discared. However, it was not considered that this may also lead to the case shown below for which the first record of the hitlets channel does not overlap with the hitlet. This leads to a wrong chopping.

• b74eae4 fixes this issue
• d3c47a4 adds some corresponding tests
• The rest is some refactoring.

The impact on the data should be minor, however a reprocessing of hitlets_nv/mv will be needed.

[WenzDaniel]

See also XENONnT/straxen#690

[JoranAngevaare]

Thanks Daniel