Add support for bfloat16

[tspiteri]

bfloat16 is another 16-bit floating point format which has the same number of exponent bits as f32, making conversion to/from f32 much faster. This PR would add support for bfloat16 floats to the crate.

I'm not sure if there is interest in adding this to the half crate, but since the code is basically a tweaked version of the f16 code, I thought it made sense to add it to the crate.

I'm also not sure about the names. I just prepended b to f16, consts, slice and vec to get bf16, bconsts, bslice and bvec. It looks fine to me, but maybe it makes more sense to keep bf16 but move the others to bfloat::consts, bfloat::slice and bfloat::vec, I don't know.