Generate assembly instructions for wordToFloat on x86_64 instead of issuing C-call. Very much inspired by what GCC does here.
Tracked in issue #22252
Instructions for source operands smaller than destination could be more efficient (e.g., 32-bit double word on x86_64),
but since Haskell source only takes Word and primop is not available in Cmm (not in Parser.y),
there is currently now way to generate this with a smaller source operand.
Note that there is a pre-existing test in testsuite/tests/codeGen/should_run/Word2Float[64|32].hs