fons points out that TH was treating 1-tuples inconsistently.  Generally
we make a 1-tuple into a no-op, so that (e) and e are the same.  But
I'd forgotten to do this for types.

It is possible to have a type with an un-saturated 1-tuple type
constructor. That now elicits an error message when converting from
TH syntax to Hs syntax