[FEATURE] Support for models that were trained on data with different sample rates

[sdatkinson]

I'd like to train models on data that with sample rates other than 48k and use them in the plugin. Currently, this could be done by hacking the training process and just knowing the sample rate to pick when running the plugin. But, I'd like this to be more fool-proof. This also currently rules out having two models with different expected sample rates from being used at the same time.

[sdatkinson]

Tracking for next release, but planning to implement #59 instead, which will make this obsolete.

[sdatkinson]

I guess this is solved by #331 and #338. #59 is the next frontier!