perf: reduce stackedmodelconfig queries

[ormsbee]

TieredCache is a hybrid request cache + regular cache, where we check
against a Django cache for the first get, but then store that data in a
process-local request cache. ConfigurationModels were already converted
to use TieredCache for performance reasons, but we never converted the
StackedConfigurationModel subclass.

This can reduce the lookups for ContentTypeGatingConfig by hundreds of
network cache calls in a given request (depending on how many exams are
in a course).

This also affects the following subclasses of StackedConfigurationModel,
though I have not measured the precise effects:

* CourseDurationLimitConfig
* DisableProgressPageStackedConfig
* DiscountRestrictionConfig
* DiscountPercentageConfig
* ProviderFilter (for dicusssions providers)
* SelfPacedRelativeDatesConfig