Adds support for FFTW3 and improved time resolution of frames.

[pjsg]

This does a number of things:

• Uses fftw3 instead of kiss by default. This is to improve CPU performance
• Uses a log10 approximation. This improves CPU performance
• Sets kTime_osr to a value such that we get output timestamps to 10ms precision. Uses optimised search process not to create lots of candidates.
• Adds the decoded hex packet to the end of the log line
• Adds compiler options (-march=native) to make it compile better on specific chipsets

The result of this is reduced CPU consumption (around halved apples-to-apples). However, since this version defaults to oversampling in time by 16 or 5 (ft8, ft4) rather than 2, it actually does a lot more work and ends up using about the same amount of CPU. It also generates maybe 5% more decodes as it aligns the ft4/8 packet better (in time).

The Makefile builds both versions of the decode_ft8 -- one using FFTW3 and one using KISS.