Investigate faster AVR floating point implementation

[nic-starke]

Currently the project uses the software floating point implementation provided by AVR GCC, however a faster implementation probably exists.

1. Check what options are available for XMEGA (or 8-bit systems).
2. Implement an option.
3. Perform basic tests to validate the performance.
4. Replace existing floating point usage with library (if it is faster).

[VegaDeftwing]

I haven't started digging through the code yet, but I can't see any good reason there'd be any floats at all? Depending on how heavily their used it may not be impractical to just rewrite anything with floats. If there's particular pain points intermediate fixed point stuff can probably be written as shims until it's de-floated.