[InstCombine] Fold exp(subnormal) and exp(+inf) to x + 1.0 (#243862)
Fixes #243132
- Folds `exp(subnormal)` to `1.0` in `SimplifyDemandedUseFPClass` since
any subnormal floating-point value evaluated with `exp`, `exp2`, or
`exp10` rounds to `1.0` in default rounding mode.
- Folds `exp(zero | subnormal | +inf | nan)` to `x + 1.0` (`fadd x,
1.0`). Note that `-inf` is excluded because `-inf + 1.0 = -inf !=
exp(-inf) = 0.0`.
- Adds test coverage across `exp`, `exp2`, and `exp10` in
`simplify-demanded-fpclass-exp.ll`.