Slow model reloads in large codebases

[kubukoz]

I've started to notice model reloads taking ~8s in a large codebase - need to identify the culprit and optimize it, I suspect some caching is missing in the schema compilers but it could be something else.