FastAI (lib and audio module) V1->V2 Migration

[pastorep]

Our aifororcas_livesystem models are trained with FastAI v1 and fastai_audio (the V1 version of an audio module built on top of FastAI v1). FastAI v2 is now the preferred version for training models. The benefits of replacing our usage of v1 with v2 (for both FastAI library and fastAI audio module) are plentiful:

• Documentation of v1 audio module is very poor (hard to debug, to find answers to questions, to understand by newcomers). Documentation of v2 audio is much better, and project is still in actively-maintained by FOSS community.
• There are lots of library version conflicts between audio v1 library and some of our other libraries (ex. librosa). Package management becomes less challenging if we move to FastAI v2 and fastaudio (the v2 version of fastai_audio).
• It'll be nice to leverage new FastAI v2 technology as we build more sophisticated models. FastAI v1 is no longer receiving non-bug-fix updates

[bnestor]

A branch is ready, but not back compatible with FastAIv1 models here: https://github.com/bnestor/aifororcas-livesystem