Navigation algos should accept non-uint8_t images as inputs

[alexdewar]

I'm not suggesting that we should use anything other than uint8_t matrices internally, but it would be good if the train/test methods transparently converted other image types rather than throwing an error. I'm currently hitting errors resulting from this in using the Python module and while I could easily do the conversion on the Python side, it would be cleaner (and maybe faster?) to do it on the C++ side.

[alexdewar]

This will also require changes to the Python wrapper code.

[alexdewar]

See: BrainsOnBoard/navbench#22