This is the first step towards schema<=>association decoupling.
Associations are now lazy-loaded via components and relations use their
own association resolver rather than relying on schema.

SchemaDSL still provides association definitions but this can be easily
improved so that Relation class can define its associations outside of
the schema block.

Associations are scoped to relations but it's possible to store them as
first-class components like the rest. I haven't done that though because
I'm not sure if this will be useful.

Here's a practical outcome of this change:

```ruby
\# here we get a relation object but its associations are NOT loaded yet
rom.relations[:users]
\#<ROM::Relation[Users] name=ROM::Relation::Name(users) dataset=#<Sequel::Postgres::Dataset: "SELECT \"users\".\"id\", \"users\".\"name\" FROM \"users\" ORDER BY \"users\".\"id\"">>

\# we can inspect the association resolver though and it still doesn't load associations
rom.relations[:users].associations.keys
["associations.users.posts", "associations.users.aliased_posts", "associations.users.labels", "associations.users.books"]

\# once something asks for a specific association, it will be lazy-loaded
rom.relations[:users].associations[:labels]
\#<ROM::SQL::Associations::ManyToMany definition=#<ROM::Associations::Definitions::ManyToMany source=ROM::Relation::Name(users) target=ROM::Relation::Name(labels) result=:many> source=#<ROM::Relation[Users] name=ROM::Relation::Name(users) dataset=#<Sequel::Postgres::Dataset: "SELECT \"users\".\"id\", \"users\".\"name\" FROM \"users\" ORDER BY \"users\".\"id\"">> target=#<ROM::Relation[Labels] name=ROM::Relation::Name(labels) dataset=#<Sequel::Postgres::Dataset: "SELECT \"labels\".\"id\", \"labels\".\"name\" FROM \"labels\" ORDER BY \"labels\".\"id\"">>>
```

There are various improvements that should be made eventually. Probably
the most important one is adding `Association#inverse_of(another)` so
that resolving "another side" can be done in a more robust way.
For example if you have "users.has_many(:posts)" but on the other side you
have "posts.belongs_to(:user, as: :author)" then it's tricky to figure it
out without a proper abstraction. Luckily, we have all the information
in the right place to implement it.