Fix logic error in PGO schema comparison

[AndyAyersMS]

I found this while trying to figure out what was leading to #51908.

I don't think this is the problem as SchemaMergesItemsWithDifferentOtherFields currently always returns false.

[AndyAyersMS]

@davidwrighton PTAL

[jakobbotsch]

The meaning seems to be flipped. If the schema merges items different other fields, then they should always be considered equal. Otherwise they should be considered equal only if the Other fields are equal. (EDIT: Or maybe I misunderstand the meaning of the function name...)

But I don't think that's the check we want anyway. It seems like this code might be causing a lot of problems with edges. Right now, if multiple .mibc files have profile data for the same method, we end up merging all edges outgoing from the same basic block into one edge. So we lose practically all edge counts.

We should really just be checking if the instrumentation kinds are edges: jakobbotsch@45bcb3f
(this also fixing merging of 64-bit and 32-bit counts at the same time). There are lots of diffs when comparing the current .mibc we consume with a new one when we do that.

[AndyAyersMS]

Or maybe I misunderstand the meaning of the function name.

I agree the name is confusing.

I think your change is clearer and also clearly improves behavior, so I'll update this PR to use that instead.

[AndyAyersMS]

Failure looks like an instance of #52710.

@davidwrighton can you should review this one?

[davidwrighton]

Long term, this code needs to be owned and understood most by the JIT team PGO experts, so if this looks good to both of you, I'm happy with whatever structure you like.