Optimize calculation/aggregate/relationship merging/reuse in calculation dependencies

We had to make a recent fix that reduces drastically the amount of cases where calculations, relationships and aggregates required by calculations in their `load/3` callback can be reused in the case that they are being loaded. This is because there is an important semantic difference: calculation dependencies should not have authorization rules applied to them.

However, there should be some analysis that can be applied to determine the ability to share, for example:

- if the calculation is being run with `calculate/3`, we can reuse it.
- If the calculation doesn't reference any aggregates or relationships, we can reuse it (traversing any nested calculations)
- If the calculation references relationships and those relationships have no authorizers

These are some naive optimizations that could be applied.

Start by tracing calculation dependency loading around the `load/3` and `calculate/3` callbacks, then inspect how authorization rules affect calculations, relationships, and aggregates. Done means reuse is safely identified for the stated cases without applying authorization rules where calculation dependencies must avoid them.

Written by the indexing model from the issue text.