Ormar

Features
* Add choices values to `OpenAPI` specs, so it looks like native `Enum` field in the result schema.

Fixes
* Fix `choices` behavior with `fastapi` usage when special fields can be not initialized yet but passed as strings etc.

Important
* **Braking Fix:** Version 0.8.0 introduced a bug that prevents generation of foreign_keys constraint in the database,
both in alembic and during creation through sqlalchemy.engine, this is fixed now.
* **THEREFORE IF YOU USE VERSION >=0.8.0 YOU ARE STRONGLY ADVISED TO UPDATE** cause despite
that most of the `ormar` functions are working your database **CREATED with ormar (or ormar + alembic)**
does not have relations and suffer from perspective of performance and data integrity.
* If you were using `ormar` to connect to existing database your performance and integrity
should be fine nevertheless you should update to reflect all future schema updates in your models.


Breaking
* **Breaking:** All foreign_keys and unique constraints now have a name so `alembic`
can identify them in db and not depend on db
* **Breaking:** During model construction if `Meta` class of the `Model` does not
include `metadata` or `database` now `ModelDefinitionError` will be raised instead of generic `AttributeError`.
* **Breaking:** `encode/databases` used for running the queries does not have a connection pool
for sqlite backend, meaning that each querry is run with a new connection and there is no way to
enable enforcing ForeignKeys constraints as those are by default turned off on every connection.
This is changed in `ormar` since >=0.9.0 and by default each sqlite3 query has `"PRAGMA foreign_keys=1;"`
run so now each sqlite3 connection by default enforces ForeignKey constraints including cascades.

Other

* Update api docs.
* Add tests for fk creation in db and for cascades in db

Features

* Introduce processing of `ForwardRef` in relations.
Now you can create self-referencing models - both `ForeignKey` and `ManyToMany` relations.
`ForwardRef` can be used both for `to` and `through` `Models`.
* Introduce the possibility to perform two **same relation** joins in one query, so to process complex relations like:

B = X = Y
//
A
\
C = X = Y <= before you could link from X to Y only once in one query
unless two different relation were used
(two relation fields with different names)

* Introduce the `paginate` method that allows to limit/offset by `page` and `page_size`.
Available for `QuerySet` and `QuerysetProxy`.

Other

* Refactoring and performance optimization in queries and joins.
* Add python 3.9 to tests and pypi setup.
* Update API docs and docs -> i.e. split of queries documentation.

Breaking
* **Breaking:** `remove()` parent from child side in reverse ForeignKey relation now requires passing a relation `name`,
as the same model can be registered multiple times and `ormar` needs to know from which relation on the parent you want to remove the child.
* **Breaking:** applying `limit` and `offset` with `select_related` is by default applied only on the main table before the join -> meaning that not the total
number of rows is limited but just number of main models (first one in the query, the one used to construct it). You can still limit all rows from db response with `limit_raw_sql=True` flag on either `limit` or `offset` (or both)
* **Breaking:** issuing `first()` now fetches the first row ordered by the primary key asc (so first one inserted (can be different for non number primary keys - i.e. alphabetical order of string))
* **Breaking:** issuing `get()` **without any filters** now fetches the first row ordered by the primary key desc (so should be last one inserted (can be different for non number primary keys - i.e. alphabetical order of string))
* **Breaking (internal):** sqlalchemy columns kept at `Meta.columns` are no longer bind to table, so you cannot get the column straight from there

Features
* Introduce **inheritance**. For now two types of inheritance are possible:
* **Mixins** - don't subclass `ormar.Model`, just define fields that are later used on different models (like `created_date` and `updated_date` on each child model), only actual models create tables, but those fields from mixins are added
* **Concrete table inheritance** - means that parent is marked as `abstract=True` in Meta class and each child has its own table with columns from the parent and own child columns, kind of similar to Mixins but parent also is a (an abstract) Model
* To read more check the docs on models -> inheritance section.
* QuerySet `first()` can be used with `prefetch_related`

Fixes
* Fix minor bug in `order_by` for primary model order bys
* Fix in `prefetch_query` for multiple related_names for the same model.
* Fix using same `related_name` on different models leading to the same related `Model` overwriting each other, now `ModelDefinitionError` is raised and you need to change the name.
* Fix `order_by` overwriting conditions when multiple joins to the same table applied.

Docs
* Split and cleanup in docs:
* Divide models section into subsections
* Divide relations section into subsections
* Divide fields section into subsections
* Add model inheritance section
* Add API (BETA) documentation

Fix issue 73

* Allow multiple relations to the same related model/table.
* Fix for wrong relation column used in many_to_many relation joins (fix 73)
* Fix for wrong relation population for m2m relations when also fk relation present for same model.
* Add check if user provide related_name if there are multiple relations to same table on one model.
* More eager cleaning of the dead weak proxy models.