check all records for endtime determination as it affects chunking

Created by: jorana

This is the straxen counterpart of https://github.com/AxFoundation/strax/pull/281. Please see the problem explained there.

The problem See https://github.com/AxFoundation/strax/pull/281.

Symptoms During a high rate run we didn't observe any problems for raw_records and records however for lone_hits we ran into:

ValueError: Attempt to create chunk [008543.raw_records: 1594281342sec 999999000 ns - 1594281348sec 499999000 ns, 5836752 items, 258.9 MB/s] whose data ends late at 1594281348499999220

This was presumably caused by the fact that while the (long) pulse was in the middle of raw_records and records but was classified as the last lone_hit leading to the error above.

Daqreader implementation As illustrated nicely below by Daniel we need to make sure we have all the latest possible wfs of each channel as raw_records may overlap. In a worse case scenario we need to look for all the channels and each of the channels has a 1 sample long pulse (which is unrealistic but lets assume this for the sake of argument).

This means we have to look back for a single channel the length of each of the samples. Next we treat all the channels known to the daqreader as an independent channel with its own timestamps (which is also very conservative). Hence we have to look back:

n_samples * length_of_samples