Replaces the use of Rodio's basic linear resampler with an fft based
resampler from the rubato crate. As we are down-sampling to the minimal
(transparent) sample rate for human speech (16kHz) any down-sampling
artifact will be noticeable.

This also refactors the rodio_ext module into sub-models as it was
getting quite long.

Release Notes:

- N/A