One approach to fixing lazy parsing of registries via from-ast

[frenchy64]

The call to -property-registry was defeating all the laziness of -delayed-registry in from-ast.

I'm not sure which way to go with this. Some options:

1. Remove call to -property-registry and allow :properties to be a map from key to IntoSchema. This works for straightforward cases because -property-registry is called again to coerce the properties back to forms, and -property-registry always calls m/schema first. But this changes the output of (-> s m/properties :registry).
2. Add more indirection in the problematic call to -property-registry to wrap schemas in a similar way to -proxy-schema, such as (-property-registry r options #(reify Schema (-validator [_] (m/schema % options)) ...)
3. Remove the use of -delayed-registry. Performance is apparently still great without laziness compared to vector syntax.

This PR implements option 1 to demonstrate the problem.

Also refactors cond into if, removing redundant tests.

[opqdonut]

I tried to understand this change.

• I couldn't really understand the context of the addition of -delayed-registry in Schema AST, Cached and other performance improvement #544 – it seems like it was a needed fix to get some recursive cases to pass
• Somehow, option 3 appeals to me, probably because I don't understand -delayed-registry
• Could this get fixed via Property Registries should allow IntoSchema instances #1181 ? Or is option 1 already fixing Property Registries should allow IntoSchema instances #1181 ?