gh-152215: Add uop peepholes for folded stack shuffles and adjacent guards

[cocolato]

This PR adds some small peephole helper functions to the cleanup phase of the
uop optimizer.

It splits the existing push/pop cleanup operations into a helper function and
adds a specific helper function for the stack reordering operations left over
from constant folding.

It also adds a helper that combines matching adjacent TOS and NOS guards
with the same side exit.

On AArch64, this allows two float guards to reuse the loaded _PyFloat_Type pointer, removing these instructions from the second
guard:

    and     x8, x24, #0xfffffffffffffffe
    ldr     x8, [x8, #0x8]
    adrp    x9, _PyFloat_Type
    ldr     x9, [x9]
    cmp     x8, x9
    b.ne    exit

    and     x8, x25, #0xfffffffffffffffe
    ldr     x8, [x8, #0x8]
    adrp    x9, _PyFloat_Type    // removed by fusion
    ldr     x9, [x9]             // removed by fusion
    cmp     x8, x9
    b.ne    exit

This reduces the stencil from 12 instructions (48 bytes) to 10 instructions
(40 bytes).

A generic stack simulator could perform more peephole optimizations, but for
now there don’t seem to be many examples that could be optimized, and it would
increase maintenance costs.

• Issue: Add effective peephole optimizer for uop traces. #152215