[HLSL] bitcasts emitted instead of vector conversions in binary operators

[llvm-beanz]

Clang currently is resolving different cast expressions for binary operator arguments than for function argument expressions. This causes bitcasts to be used instead of correct implicit conversion sequences.

See reproduction case: https://hlsl.godbolt.org/z/jKGcGbE4Y

This change will require the following:

• Document Usual Arithmetic Conversions microsoft/hlsl-specs#308
• Disabling "Lax" vector conversions in Clang
• Implementing the spec as defined in the spec from the issue above