Note [No eta reduction needed in rhsToBody]

Historical note.  In the olden days we used to have a Prep-specific
eta-reduction step in rhsToBody:
  rhsToBody expr@(Lam {})
    | Just no_lam_result <- tryEtaReducePrep bndrs body
    = return (emptyFloats, no_lam_result)

The goal was to reduce
        case x of { p -> \xs. map f xs }
    ==> case x of { p -> map f }

to avoid allocating a lambda.  Of course, we'd allocate a PAP
instead, which is hardly better, but that's the way it was.

Now we simply don't bother with this. It doesn't seem to be a win,
and it's extra work.