inline loops that we know will only execute once

[superlopuh]

This should help us run the conv and pooling kernels as they have multiple 1-dim loops each.

MWE(?) as filecheck test:

// RUN: xdsl-opt -p canonicalize %s | filecheck %s

riscv_func.func @at_least_once() {
    %one = riscv.li 1 : () -> !riscv.reg<>
    %three = riscv.li 3 : () -> !riscv.reg<>
    %0 = riscv.mv %one : (!riscv.reg<>) -> !riscv.reg<>
    riscv_cf.bge %0 : !riscv.reg<>, %three : !riscv.reg<>, ^0(%0 : !riscv.reg<>), ^1(%0 : !riscv.reg<>)
^1(%i : !riscv.reg<>):
    riscv.label "scf_body_0_for"
    "test.op"(%i) : (!riscv.reg<>) -> ()
    %1 = riscv.add %i, %one : (!riscv.reg<>, !riscv.reg<>) -> !riscv.reg<>
    riscv_cf.blt %1 : !riscv.reg<>, %three : !riscv.reg<>, ^1(%1 : !riscv.reg<>), ^0(%1 : !riscv.reg<>)
^0(%2 : !riscv.reg<>):
    riscv.label "scf_body_end_0_for"
    riscv_func.return
}

// CHECK-NEXT:    riscv_func.func @at_least_once() {
// CHECK-NEXT:      %one = riscv.li 1 : () -> !riscv.reg<>
// CHECK-NEXT:      %three = riscv.li 3 : () -> !riscv.reg<>
// CHECK-NEXT:      %0 = riscv.mv %one : (!riscv.reg<>) -> !riscv.reg<>
// CHECK-NEXT:    riscv_cf.branch ^1(%0 : !riscv.reg<>)
// CHECK-NEXT:    ^1(%i : !riscv.reg<>):
// CHECK-NEXT:      riscv.label "scf_body_0_for"
// CHECK-NEXT:      "test.op"(%i) : (!riscv.reg<>) -> ()
// CHECK-NEXT:      %1 = riscv.add %i, %one : (!riscv.reg<>, !riscv.reg<>) -> !riscv.reg<>
// CHECK-NEXT:      riscv_cf.blt %1 : !riscv.reg<>, %three : !riscv.reg<>, ^1(%1 : !riscv.reg<>), ^0(%1 : !riscv.reg<>)
// CHECK-NEXT:    ^0(%2 : !riscv.reg<>):
// CHECK-NEXT:      riscv.label "scf_body_end_0_for"
// CHECK-NEXT:      riscv_func.return
// CHECK-NEXT:    }