Add global cache for available resource types

Currently the resource types will be loaded every time a semantic model is built. To improve compilation performance, we should implement a global cache for resource types.

Start by locating where resource types are loaded during semantic model construction and trace how often that path runs. Implementing the global cache should prevent repeated loads across model builds; confirm this with compilation or performance tests, though the issue does not name specific files or tests.

Written by the indexing model from the issue text.