New codegen for inline array fields

[jnm2]

Closes #117

TODO:

• Add range checks to the fallback indexer implementation to parallel span range checks
• Ask JaredPar to review the generated code and see if ref p[index] should be ref Unsafe.AsRef<uint>(p + index)

Currently generated code (odd whitespace is not my fault; tracked by #23):

[StructLayout(LayoutKind.Sequential, Pack = 1)]
internal struct __dwReserved_4
{
    internal uint _1, _2, _3, _4;
    internal ref uint this[int index]
    {
        [MethodImpl(MethodImplOptions.AggressiveInlining)]
        get
        {
            unsafe
            {
                fixed (uint *p = &_1)
                    return ref p[index];
            }
        }
    }
}

@AArnott I like keeping unsafe scoped as tightly as possible, meaning I use the statement form if the signature itself isn't unsafe, but let me know if this is too weird or something, or if using an embedded statement instead of a block inside the fixed statement is a problem.

[AArnott]

@tannergooding were you saying we don't need to specify the pack size?

[tannergooding]

Right. Inline arrays are just like regular arrays and the natural padding of a struct is correct.

[tannergooding]

(assuming the metadata doesn't specify otherwise)

[jnm2]

Saw that too. I can add this change to this PR, no problem. 👍

[AArnott]

I've asked JaredPar on another thread to review and will report back here.

[AArnott]

Just an update: I am actively working on this and have lots of changes coming. I'm just working to get all tests passing. The new extension methods we have to declare bring on some new challenges.

[jnm2]

@AArnott Thanks for completing this!

[AArnott]

Thanks for starting it, @jnm2! Particularly because it led us to discover bugs in how we did it before.