Define inline asm consistently in intel syntax with -masm=intel

[pedrolamarao]

One of the greatest pains in reading x86 assembler code is deducing what syntax is going on, AT&T or Intel. This is not a simple matter of style: the order of operands is reversed, making simple extended asm statements such as mov cr3, %0 ambiguous.

Recently released clang 14 allows defining the syntax for inline asm with -masm=syntax, like gcc does. Let's use that to normalize all inline asm to Intel syntax.

In case we return to assembler source files, .intel_syntax is already supported.

[pedrolamarao]

This obviously upgrades LLVM requirement to 14.