Use cached Malli validators/explainers; make `:fn` schemas cacheable (#39947)

This improves our Malli performance substantially! Many of our code paths were calling vanilla `malli.core/validate` or `malli.core/explain`; these redo the (possibly expensive) compilation of the schema into a validator or explainer for every call.

We should use the caching versions in `metabase.util.malli.registry` everywhere, and our runtime performance will be much improved.

However when I started using these cached versions, I found that the memory use was growing out of control. Eventually I tracked this down to `:fn` schemas. Functions are only comparable by pointer equality, so they make poor cache keys. `(fn ...)` or `comp` calls in a schema on a `mu/defn` function **get re-created for every call of the `mu/defn`'d function**! That's a big time sink recompiling the schemas if we're *not* caching, and a huge memory sink if we *are* caching!

This PR pulls every such schema I could find out into a `def` so it uses the same closure and is cacheable. I'd like to automate that in some part of the Malli pipeline, or maybe a linter rule, but I haven't found a good way to do it yet.

Part of #39946.