Adapt to DLamCasesE in th-desugar-1.18

[RyanGlScott]

This patch bumps the pinned th-desugar commit to bring in the changes from goldfirere/th-desugar#218, which replaces DLamE and DCaseE in favor of DLamCasesE. Quite happily, this actually simplifies how singling works in singletons-th. Previously, we went through great efforts to annotate promoted case expressions with their overall return type, as described in Note [Annotate case return type] and Note [The id hack; or, how singletons-th learned to stop worrying and avoid kind generalization] in D.S.TH.Single. After this patch, however, all case expressions are desugared to \cases expressions, and because we already annotate singled \cases expressions (by generating code like singFun1 @LamCasesSym0 (\cases ...)), we no longer need to use the tricks describes in those two Notes. Hooray!

One interesting knock-on effect of these simplifications is that given code like this:

map (\x -> ()) xs -- Note that `x` is unused

singletons-th would previously generate code that looked like this:

sMap (singFun1 @LamdaSym0 (\sX -> case sX of (_ :: Sing x) -> STuple0)) sXs

Unlike the original code, this singled code would compile without warnings. This is because sX is "used" in the sense that it is passed to a case expression. It's a very silly case expression, as it doesn't do anything meaningful with sX, but it still technically counts as a use.

Now that singletons-th uses a \cases-based approach to singling, however, the singled code will instead look like this:

sMap (singFun1 @LamCasesSym0 (\cases (sX :: Sing x) -> STuple0)) sXs

This time, GHC does recognize that sX is unused and emits a warning. This actually affects the way that singletons-th generates code for derived Functor and Foldable instances, so I needed to generate wildcard patterns instead of variable patterns in certain parts of
D.S.TH.Promote.Deriving.{Functor,Foldable}. I have also mentioned the possibility of singletons-th generating more warnings in the CHANGELOG.