1) dunno, i always used CR2 under the theory that nothing is changing the data if you do that.
2) no; flats, darks and biases need to be applied to the lights in raw, undebayered format
3) sort of, usually with a mono CCD you have L, R, G and B images. you can make a 'fake' L channel out of your RGB data but it's not the same as a real Luminance image.