Should handle the history in a more consistent way.
Slightly increase the model size and re-enable biased band loss in
training.

Starting with compute_linear()

Shift computation to concealment

Should make synthesis easier in the future

We're back to processing 10 ms at a time and have no need for 40-ms
"superframes".

Calls are deferred to the actual loss and we only process the minimum
required.

Can now inject FEC features to be used by the PLC when available

Helps forget concealment features faster

Variable cannot be uninitialized

Causal vs non-causal, DC filter, and blending are now configurable