Use TAdd in the instance head instead of a TSub context in TupleSize

[krame505]

This reverts commit 9d77afc.

This was a change I apparently made in #729, for reasons that I don't recall. However it turns out to cause a performance regression in our code base on a module with a huge number of input ports of complex types.

[quark17]

Yeah, I avoid TSub because the undefined cases make it not entirely compatible with TAdd. Though, in this case, is it the use of a type function in the head that improves the performance? That is, would this equally perform poorly?:

instance (TupleSize' b n, Add n 1 m) => TupleSize' (a, b) m where {}

[krame505]

I didn't try that variant, but I suspect that the difference is TAdd resolves during matching without creating an additional context to solve.

[quark17]

But then, why doesn't TSub also resolve without needing an additional context?