[MIR] Support +inf and -inf literals (#221127)
Support parsing `+inf` and `-inf` as floating-point literals in MIR.
The MIR lexer previously tokenized the leading sign separately, causing
`G_FCONSTANT float +inf` and `G_FCONSTANT float -inf` to fail with
"expected a floating point literal".
Recognize signed infinity spellings as `FloatingPointLiteral` tokens so
they can be handled by the existing floating-point constant parsing
path.
Add a regression test covering both positive and negative infinity.
Fixes #221040
Assisted-by: ChatGPT (rubber ducking)
