incomplete l2 cache and consecutive datasets

[j-haacker]

Describe the bug
If one builds a l3 dataset for a time period that ends before the available ground tracks are exhausted, extending the dataset at a later stage results in skipping data of the month after the first l3 dataset ended.

This is because of a "dirty" convention to define the period that should be considered. The convention is to stop on the first day of the "end" time step if this is a month or a year. This results in some data in the month after the requested period being present - which, in turn, leads to that month being skipped if the datasets is being extended.

To Reproduce
Steps to reproduce the behavior:

1. l3.build_dataset("01-01", 2010, 2015)
2. l3.build_dataset("01-01", 2010, 2025, reprocess=False) # not sure about reprocess, but turning it of reproduces the behavior in any case

Expected behavior
There should never be data gaps that could even go unnoticed.