Optimise hyperparams: Medoid radius and default radius

[jakobnissen]

These parameters, defined as constants at the top of vamb/cluster.py, was changed in #204, but I'm not sure the change was benchmarked properly. These needs to be carefully optimised to get the best results.

Also note that this change slows down clustering significantly (4x slower in tests), see #250, so performance needs to be measured against time

[jakobnissen]

@Las02 : A larger MEDOID_RADIUS will cause Vamb's clustering to run slower, but may produce more accurate clusters. Please test an array of values between 0.05 (current) and 0.1. I'm not comfortable letting it be lower than 0.05, even if it's faster. Measure NC bins and time consumed during the clustering step (can be extracted from the log file)

The DEFAULT_RADIUS parameter is maybe more interesting. Changing this should have less effect on the runtime, but may have a significant effect on number of NCs... Perhaps. This too, almost certainly needs to be increased.