Commit message:
cmd/compile/internal/ssa: constant fold math/bits functions
Add constant folding rewrite rules for RotateLeft64/32/16/8,
Sub64borrow, and Div128u generic SSA operations when their inputs
are compile-time constants.
Currently, these operations are evaluated at runtime by target-specific
instruction sequences, even when all their arguments are constant.
This change leverages compiler-level optimization to evaluate these
operations statically during the generic SSA optimization pass, replacing
them with simple constant loads.
Add corresponding codegen assertions to test/codegen/mathbits.go
to verify instruction elimination across major CPU architectures
(amd64, arm64, ppc64x, loong64, riscv64, s390x, and WebAssembly).
Fixes #21872.