Wrap derived schema in Exported wrapper

[darl]

This PR fights with issue described on this issues:

• [Circe] Unclear implicit prioritisation rules softwaremill/magnolia#89
• Best way to deal with ambiguous implicit values with other libraries w.r.t. List, Option, etc? softwaremill/magnolia#213
• Separate exported instance type and internal dispatch/combine type softwaremill/magnolia#107

Short version: macro derivation sometimes conflicts with builtin schemas defined in Schema companion object.

On this PR I followed solution described by @neko-kai on this comment.

In first commit I added failing test so it can be tested it in isolation.

[ghostdogpr]

Interesting! I have some suggestion to reduce breaking changes.

[ghostdogpr]

How about calling this differently and make gen return the same thing as before? So that we don't need to change every existing call.

[darl]

Have some experiments with this.

1. Old method

implicit def gen[T]: Typeclass[T] = Magnolia.gen[T]

Will pollute implicit scope when imported.

2. Without implicit

def gen[T]: Typeclass[T] = Magnolia.gen[T]

Different semantic: magnolia disables recursive generation without implicit keyword
https://github.com/propensive/magnolia/blob/master/core/shared/src/main/scala/magnolia.scala#L249-L251

3. Using implicit

def gen[T](implicit e: Exported[Typeclass[T]]) = e.instance

Works with default schema (Schema.gen[MyClass]) but not work with custom schemas like this:

object mySchema extends GenericSchema[Blocking]
mySchema.gen[MyClass] // this will fail 

I will experiment with this a bit more on weekend.

[ghostdogpr]

What if you renamed the implicit gen into e.g. autogen and added a new def gen that is not implicit and just a shortcut to avoid calling .instance?

[ghostdogpr]

I think the 3rd one is acceptable, actually the caliban docs say you should import mySchema._ when you use this pattern, which makes the implicit in scope. I opened #658 to get this in.

[ghostdogpr]

Closing since #658 was merged