EoU classifier: better handling of sampling rate

[rfejgin]

When accepting NumPy arrays as input, its sampling rate was not defined and it was ambiguous as to whether it's the caller's responsibility to resample it to 16 kHz (for audio file inputs, the classifier was doing the resampling).

Clarify this by requiring a sampling rate parameter when providing NumPy inputs. We then resample internally to 16 kHz if needed. So now the inputs can be any sampling rate in both cases - WAV file input or NumPy input.

Also:

• removed the sampling rate argument from the classifier's constructor as it is internally always 16 kHz.
• added associated tests