Features should return consistent types

[adrien-berchet]

The NeuroM.features.* functions return inconsistent types. For example, segment_taper_rates returns a list of floats while section_taper_rates returns a list of numpy.float32. This can be annoying to process results (e.g. it's harder to dump the results to JSON strings or stuff like that).
I suggest we make all the types consistent, like casting all the results to built-in Python types.
WDYT @eleftherioszisis @mgeplf ? If you agree I can work on this.

[eleftherioszisis]

Yeah, makes sense to me to have consistent types.

[adrien-berchet]

Will be fixed in v4