Fixing store_data_start in recursive peak splitter calls and peaklet["length'] fix in store_downsampled_waveform

[HenningSE]

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

This PR fixes some smaller problems related to the data_start handling in the peaklets building/ splitting.

• The flag store_data_start was not passed to recursive calls of the peak splitter. As a result some peaklets had data_start always set to 0.
• peaklet["length'] is used in store_downsampled_waveform to select the part of the waveform that is stored in data_start. If the waveform is downsampled, peaklet["length'] is overwritten. When saving data_start we need to take the initial length of the peaklet not the downsampled length.

[coveralls]

coverage: 90.161% (+0.002%) from 90.159%
when pulling d5b02d7 on bugfixing_data_start
into 0e7f9c6 on master.

[dachengx]

@HenningSE Please make the title of this PR more detailed.