Add new opcode for moving the stack pointer to a reg and use it to eliminate custom Linux startup code

[robertmuth]

The startup code for a32 looks like this:

    unit.FunStart("_start", 16, NOP_BYTES)
    for mnemonic, ops in [
                          ("ldr_imm_add", "al r0 sp 0"),
                          ("add_imm", "al r1 sp 4"),
                          ("bl", "al expr:call:main"),
                          ("movw", "al r7 1"),
                          ("svc", "al 0"),
                          ("ud2", "al")]:
        HandleOpcode(mnemonic, ops.split(), unit)
        unit.FunEnd()

If we add an opcode get_sp, we could express this directly in Cwerg IR:

    get_sp  sp:A32
    ld  argc:U32 sp 0
    lea  argv:A32 sp 4
    push argv
    push argc
    bsr main
    poparg res:S32
    pusharg res
    bsr exit
    trap

[robertmuth]

GETFP and GETSP were added recently to address this