Scikit-digital-health

Major changes:

- Add new c/fortran extensions for features
- Refactor core features to allow for n-dimensional array inputs

Changes

- [Gait] Change CWT scale calculation to better match paper

Major Changes:
- [Gait] gait processing now always downsamples when required. Downsampling used to be only for classification, now it is for the whole process
- [Gait] added back number of valid full gait cycles to output metrics

Changes
- [Gait] Changed how gait predictions were broadcast to samples, and gait bout start/end creation at native frequency
- [Gait] Add option to save gait predictions to a file

Major Changes:
- Add a GT3X reader for Actigraph files
- [Gait] Change classification downsampling to linear from cubic, and use NumPy.interp for huge memory savings
- [Gait] Allow no downsampling for gait classification if sampling frequency is within +- 1.5% of goal frequencies
- [Gait] Change sampling period computation to use entire time series to minimize effect of poor spacings

Minor Changes:
- Add testing for edge cases for several gait metrics
- Incorporate logging and logging tests
- Harmonic ratio needs a minimum of 11 harmonics

Major Changes:
- Added logging in places where user input/code can't be changed.
- Added logging messages when starting processing, if the user is using logging

Gait changes:
- move vertical axis computation to a per-bout level to deal with potential sensor placement changes
- add a check and minimum of 10 harmonics necessary to compute the Harmonic Ratio
- removed unnecessary keys from the gait results dictionary (IC, FC, FC opp foot, b valid cycle)
- gait results returned now only include the cycles which were valid