Improve Opus codec performance by building with optimisations

derf has suggested to build Opus with the following options:

```
--enable-intrinsics --with-NE10= and possibly --enable-fixed-point
```

This means that we need to add an NE10 recipe ( https://github.com/projectNe10/Ne10/blob/master/doc/BuildingNe10.txt ) to cerbero for armv7 and arm64 Android and iOS. And then I guess test with fixed point or floating point is faster.

JonLennox from the #opus IRC channel and who has worked on some further optimisations for both floating and fixed point suggests that fixed point is faster.

Start with cerbero and the linked NE10 BuildingNe10 instructions. Determine how to add an NE10 recipe for armv7 and arm64 on Android and iOS, then compare Opus fixed-point and floating-point builds using the proposed intrinsic and optimisation options. Done means the recipe builds on the named targets and the faster configuration is established.

Written by the indexing model from the issue text.