Support `+inf` and `-inf` literals in MIR parser.

Currently, `MIR` does not support `+inf` and `-inf` literals.
```
# okay
G_FCONSTANT float 0x7FF0000000000000
G_FCONSTANT float f0x7F800000

# rejected
G_FCONSTANT float +inf
```

As noted here:
https://github.com/llvm/llvm-project/pull/221019#discussion_r3928024352

Start at the MIR parser and trace how floating-point literals are currently recognized, using the accepted hexadecimal examples as a reference. Add support for both `+inf` and `-inf`, then verify that these literals are accepted while the existing forms continue to work.

Written by the indexing model from the issue text.