Use `raiseload` option for all queries

[di]

Sort of towards https://github.com/pypi/warehouse/issues/3081.

This PR turns on the raiseload option by default for all queries, essentially making N+1 queries impossible in our codebase, and requiring explicit loads instead, rather than per-query as documented in https://docs.sqlalchemy.org/en/20/orm/queryguide/relationships.html#preventing-unwanted-lazy-loads-using-raiseload .

Marking this as draft for now because there are a lot of test failures to address. I'm also not entirely sure yet if this also covers deferred columns as well, because that seems to be a separate setting: https://docs.sqlalchemy.org/en/20/orm/queryguide/columns.html#using-raiseload-to-prevent-deferred-column-loads

[ewdurbin]

Question: Is there a way to approach this gradually? Perhaps via Warnings?

Request: Can you implement a couple of the necessary fixes as examples? You mentioned that there might be some impacts on code readability. If its too egregious it may not be worth it.

[miketheman]

Let's pursue this once we've upgraded db libs. see: #13497

[di]

@miketheman Do you think we're ready to revisit this?

[miketheman]

@di I think this is worthwhile, but should probably come after #15634 - since I was trying to ensure a query count before/after to support this effort from a web-side, since many of our potential N+1 come from HTML template rendering iteration loops.

[di]

OK! Let's say this is blocked on https://github.com/pypi/warehouse/issues/15634.

[miketheman]

OK! Let's say this is blocked on #15634.

unblocked now that we can count queries emitted when loading a page via webtest

[di]

Rebased. Quite a number of test failures to work through here if anyone is feeling up for it 🙂