We should introduce a well-tested index fingerprinting library that, given a join graph, will iterate over each node and do:
1. Extracting all possible indexes that are eligible for INLJ (if any)
2. Determine the exact set of relevant indexes
3. Hash the required fields of the relevant indexes

We should combine the hashes generated from each node in the join graph to create a single index fingerprint that we store in the JoinPlanCache for a given cache entry.