Currently, we encode the index name for INLJ nodes in toCachedJoinPlan and after SERVER-130368, will also encode it as part of the index fingerprint. This requires us to reason about situations such as:
index name is “foo” and key spec is {a: 1}, then drop and recreate index “foo” with key spec {b: 1}. Currently, we can reason through and see that there are safe guards for this but the uniquely generated ident would likely provide a better level of safety.