Fix BOX import to not create "wrong" casts

[SingleAccretion]

All casts in the IR have TYP_INT, except for this one case. Fix it by using genActualType.

Also add a comment explaining the reason a cast is needed at all in case the types are small.

No diffs as per SPMI. Note: it looks like the Enum.HasFlag optimization has been fixed to use actual types.

Fixes #10843

[SingleAccretion]

The assert was redundant with the one above the if.

[hughbe]

It would be good to have a test with this change

[SingleAccretion]

What kind do you have in mind? This change is not fixing a bug per se, just an inconsistency internal to the IR.

[hughbe]

I was thinking having tests that make sure HasFlag still produces correct results for Enums of all different sizes (unsigned/signed, 8/16/32/64 bit), e.g.

enum ShortEnum : short
{
    None,
    Boo
}

static bool Test(ShortEnum s) => s.HasFlag(ShortEnum.Boo);

[SingleAccretion]

I see. We already have a test for the optimization here, though it deals more with the intricacies of how many flavors of boxes can one see. I will see about adding some more coverage of small type mixes, though I do not expect us to have issues in this area specifically (there are definitely issues with dealing with these implicit truncations in general...).

[AndyAyersMS]

@JulieLeeMSFT you can cross this one off the list I gave you the other day.

[AndyAyersMS]

LGTM.