Optimize SQL expression comparison by caching hash codes (currently probably not needed)

In the query pipeline sometimes expressions are compared for (deep) equality.
This currently is based on a recursive visit of the subtree, which can be very costly if done multiple times while visiting the expression (it is quite easy to construct expression trees that require O(n^2) operations when visited).

This could be improved by computing a hash that filters out most of the inequalities as suggested in https://github.com/dotnet/efcore/pull/34133#discussion_r1663964197