[MPS] Enable more bfloat16 ops (#119738)
Introduce conveninence inlinable `mps::supportedFloatingType` function
that returns true if type is Float, Half or BFloat16
Test by running LLM inference using bfloat16
Pull Request resolved: https://github.com/pytorch/pytorch/pull/119738
Approved by: https://github.com/Skylion007
