SDS Plugin

[obsmax]

Dear PhaseNet developers,

My team and I are using Phasenet and were impressed by the potential of this program.
In this PR, we would like to submit a plugin to call PhaseNet in prediction mode on a large SDS data structure.
The packing of 3 components waveforms into single files can be difficult to implement for very large datasets.
With this plugin, we propose to access the 3-component waveforms directly as they are stored in a SDS (seiscomp like) data structure.

• The idea of submitting a plugin is to make sure that the changes do not interfere with the other components of the program.
• The plugin is not yet interfaced in the main run.py program, but could be quite easily under the flag --input_sds.
  To lauch it, it is enough to execute the sds_plugin.py file, or reproduce the main section of the plugin in a separate python script.
• The plugin does not require new packages, and runs under the python environment of the project
• An anonymous dataset has been added to the "demo" directory to test this plugin

input

• a csv file listing the stations and time windows to explore, see demo/fname_sds.csv

output

• a picks.csv file with picks provided in absolute time (not sample indexs)
• optionally: a hdf5 archive with the prediction time series organized in a way similar to the SDS structure
• It is also possible to generate png figures for visual QC of the picks together with the waveforms and probablity time series.