New stuff

Spectrograms are now paged (stored separately in folders not exceeding 1000 spectrograms each) so the load time is vastly improved
Features can now be aggregated by min, max, var, begin and end values
Added parameters to database model which can be set at the creation of the database, or changed in database management
Change feature extraction to use database's specific NFFT and overlap
Spectrogram preferences (colour map, zoom, etc...) are now saved - no need to change them each time you open a new file
Spectrogram now has y-axis
Added script to export koe syllables to Excel, including spectrogram
Apply bandpass filter in feature extraction
Make Koe compatible with Python 3.9, Mac Big Sur, and Apple Silicon (must use conda-forge)
Koe's docker images are now truly handsfree, no more messing with project settings on-site

Fixed stuff

Fixed various problems with spectrograms not being generated when importing CSV
Fix inconsistency in file's sampling rate across: original WAV, converted MP3 and Browser's default sampling rate
Use 2 decimal points for segment start and end times
Fix bug: end_ms > duration
Fix various bugs when upload raw recording
Fix fundamental frequency not returning correct value when the database has bandpass filter