Faster color conversions

[rfuest]

This PR changes the implementation of the color conversion to be more efficient when no hardware divider is available. The division is replaced by a cheap right shift. It is as fast as the LUT based implementation mentioned in #650, but doesn't need to store any tables in flash.

[jamwaffles]

Tiny nit, otherwise looks good. Nice use of all that const magic!

I went to run a benchmark for this and realised we don't have any for the colour conversion stuff. It shouldn't block this PR, but perhaps worth considering now or later? At any rate, how does it compare to the timings in #650?

[rfuest]

I went to run a benchmark for this and realised we don't have any for the colour conversion stuff. It shouldn't block this PR, but perhaps worth considering now or later?

IMO this would be something that needs to run on real target hardware. A beefy desktop CPU won't have a big problem dealing with some additional divisions.

At any rate, how does it compare to the timings in #650?

On my Cortex M0 test setup the benchmark result is within a few milliseconds of the LUT based implementation. But the LUT based approach might have advantages on platforms with less than 32 bit word width, no hardware multiplier and/or no barrel shifter. At the moment I think the color conversions are now fast enough until someone reports otherwise.

[jamwaffles]

True, although it would have been interesting to see any deltas if they were somewhat representative of other HW. At any rate, 50% is a pretty good improvement!