No longer needed now that PLC is trained with PyTorch stack

Used for DNN matrix multiplies

Starting with compute_linear()

Remove old pitch estimator and retrain all models

Totally untested -- most likely doesn't work

Untested

And missing prototypes

Untested

Should be able to handle all previous GRU variants and more.

first attempt of C implementation of fec encoder (not tested yet due to NEON/DOT_PROD not being separable)

Only on the C side, no sparse GRU B training yet