Understand differences with off-line analysis

The correlation between lstchain Hillas and timing parameters and the SAG-RECO's parameters is not perfect, even though almost all algorithm are the same. We need to understand the differences to search for bugs, or configuration improvements.

Story list:

• - understand impact of less robust gradient computation algorithm
• - understand differences when waveform peak is near the window border
• 47e55dec - apply "static" (measured from flat field events) time calibration
• - read from data stream and apply DRS4 time shift correction, see https://gitlab.cta-observatory.org/cta-computing/data-model/data-model-group/-/issues/67