Summary

This PR extends BoxIterator to be n-dimensional. Additionally, it now supports a range-based for loops. The hope is that this can be useful to reduce the reliance on ifdefs/macros to write loops over boxes that run on CPUs, though so far I haven't found that many places where this can be used.

Box b;
...
for (IntVect iv : b.iterator()) {
    // do operations involving iv
}

Note that while regular iteration of the range-based version is reasonably performant,
it is not compatible with omp parallel for or simd autovectorization.

Additional background

The range-based version produces good assembly when compiling with -O3, having minimal instructions in the innermost loop.
https://godbolt.org/z/MTcM8Ysaq

void g (const IntVect&);

void test (Box bx) {
    for (auto iv : bx.iterator()) {
        g(iv);
    }
}

        ...
.L3:
        mov     ebx, DWORD PTR [rsp+8]
.L4:
        mov     r15d, r13d
.L6:
        mov     DWORD PTR [rsp+16], r15d
        lea     rdi, [rsp+16]
        add     r15d, 1
        mov     DWORD PTR [rsp+20], ebx
        mov     DWORD PTR [rsp+24], ebp
        call    "g(IntVectND<3> const&)"
        cmp     r15d, r12d
        jle     .L6
        add     ebx, 1
        cmp     r14d, ebx
        jge     .L4
        cmp     ebp, DWORD PTR [rsp+12]
        je      .L13
        add     ebp, 1
        jmp     .L3
        ...

Checklist

The proposed changes:

• fix a bug or incorrect behavior in AMReX
• add new capabilities to AMReX
• changes answers in the test suite to more than roundoff level
• are likely to significantly affect the results of downstream AMReX users
• include documentation in the code and/or rst files, if appropriate