Add Whisper speech recognition example

[robertknight]

Add a speech recognition example using Whisper.

This uses Hugging Face's implementation for consistency with other examples. The upside is that HF maintain these models and their docs. The downside is that the Python model code is much more complicated to dig into than OpenAI's original implementation or other ONNX-exportable implementations.

• Add Whisper example that has been tested with the Tiny, Base, Small and Medium V3 models. The Large Turbo model doesn't work, for reasons yet to be determined.
• Add script to export mel weight matrices from librosa. This is the same procedure that was used to generate the mel weight matrices in OpenAI's original Whisper implementation.
• Add microfft dependency for FFT of f32 signals. This is a small and portable FFT implementation.

Performance is OK, but there is an expensive Transpose operation showing up in profiles, especially for the larger models, which needs looking into.