Access permissions 5: DRY permission caching

[lukasbestle]

🚨 Merge first

• Access permissions 3: Smaller refactorings #6879
• Access permissions 4: Fix user caching #6880

Description

Summary of changes

Move caching of access and list permissions to ModelPermissions

Reasoning

DRY code, especially now that we would add five more of these methods to the User, Site and Language classes

Additional context

We could cache all other permissions as well, but I've wrestled with unit tests for too long before Christmas. The write permissions would also benefit less from the caching as there is usually only one write operation per request (in contrast to potentially hundreds of list/access permission checks).

Another advantage of the separate canFromCache() method is that it makes explicit which checks come from cache and which are dynamic. If we ever use canFromCache() in a place that needs dynamic checking, we will immediately be able to tell because of the LogicException that is thrown then.

The new UserPermissions::cacheKey() is currently uncovered as the User class does not have access/list permissions at the moment. This will be fixed in the PR 7.

Changelog

Refactoring

• Globally cache access and list permissions per permission category, model type and user role to reduce code duplication

Breaking changes

None

Docs

None

Ready?

• In-code documentation (wherever needed)
• Unit tests for fixed bug/feature
• Tests and CI checks all pass

For review team

• Add lab and/or sandbox examples (wherever helpful)
• Add changes & docs to release notes draft in Notion

[lukasbestle]

I think this PR really does have a performance impact, see #6880 (comment). With these changes, we need to create a ModelPermissions instance for each model and not just for each combination of template and user role.

I wonder if there is a better way to DRY the logic without this performance regression. Or otherwise we should probably discard this and add the same caching logic to the new isAccessible()/isListable() methods for site/user/language in upcoming PR 7.

The performance impact should now be fixed with a different approach: I moved the static caching logic to a static method inside ModelPermissions. This effectively does the same as before, just in one place to avoid the code duplication.