Using a max lambda of 0.04

Remove old pitch estimator and retrain all models

More efficient than PVQ

There was a suggestion to use 127 for extending the extension ID space

When ready, change DRED_EXTENSION_ID to the final ID, remove
DRED_EXPERIMENTAL_VERSION completely, and change DRED_EXPERIMENTAL_BYTES
to zero (eventually remove it).

The bits we don't use won't be wasted, so it's less important to
get exactly the optimal number of bits below the cap.

The gain*2 when overshooting was too aggressive and the undershoot
case wasn't aggressive enough. This now seems to work reasonably well.

Don't attempt to run the neural PLC on the side channel since
we only have one state.

Also, don't code DRED that's redundant with the main packet

Should make synthesis easier in the future

8k, 12k and stereo are mostly untested

So it can be used outside of the SILK PLC

Brings the DRED implementation in line with the -00 draft

output() renamed to decode(), dred objects using alloc() and free(),
OpusDRED now passed as cost for decoding.

Everything can now be shallow-copied again

Makes shallow copy of decoder possible again