Take care to not eta-reduce jumps in CorePrep
In #17429 (closed) a bug came up where CorePrep was too eager to eta-reduce
uses of JoinIds
, leading to a Lint failure.
This patch fixes that issue by making tryEtaReducePrep
consider
JoinId
s not OK-for-eta-reduction in the same way it treats Id
s that
respond True to hasNoBinding
.