Massive performance update, dumping data to ids and not nested models and other

0.25.0

⚡ Performance

A broad performance pass across model construction, row hydration, relation
resolution, alias lookups, and result merging — combining per-class caching,
lazy initialization of relation machinery, and porting the hottest helpers to
Rust via the new required ormar-utils
dependency. End-to-end speedups vs 0.24.0 range from ~12 % on select_related,
through ~30 % on iterate and first, to ~60 - 80 % on initialization of models, and up to x2.5 speedup in getting all.

✨ Features

• Add flatten_fields() queryset method and model_dump(flatten_fields=...) /
  model_dump(flatten_all=True) arguments to render related models as their
  primary key value instead of nested dicts — see docs #1641

  cars = await Car.objects.flatten_fields("manufacturer").all()
  assert cars[0].model_dump() == {"id": 1, "name": "Corolla", "manufacturer": 1} # Note the id 1 and not the nested model
  
  car.model_dump(flatten_all=True)
  car.model_dump(flatten_fields={"manufacturer": ...})

• Add proxy model inheritance (OrmarConfig(proxy=True)) — child class
  shares the parent's table and only adds methods / computed fields / a custom
  queryset_class — thanks @SepehrBazyar — see docs #760

  class User(Human):
      ormar_config = base_ormar_config.copy(proxy=True)
  
      def full_name(self) -> str:
          return f"{self.first_name} {self.last_name}"

• Add per-table schema= support on OrmarConfig so models can live in
  non-default database schemas (PostgreSQL full, MySQL with caveats, SQLite
  not supported for cross-schema FKs) — thanks @hk3dva — see docs #1493

• Add Python-style indexing and slicing on QuerySet — Track.objects[:10],
  Track.objects[-5:], Track.objects[5] — thanks @SepehrBazyar — see docs #765

• Add nulls_ordering argument to .asc() / .desc() (ormar.NullsOrdering.FIRST
  / LAST) with MySQL emulation — thanks @SepehrBazyar — see docs #769

• Allow filtering by a foreign-key column directly (Book.objects.filter(Book.author == 5)
  or with a model instance) without adding a JOIN — thanks @djalouehz — see docs #854

• Add through_relation_nullable and through_reverse_relation_nullable to
  ormar.ManyToMany to enforce NOT NULL on the through table's FK columns —
  thanks @evadev-eva — see docs #852

• Add last() / last_or_none() on QuerySet and expose first_or_none() /
  last() / last_or_none() on QuerysetProxy — see docs #765

• Forward unrecognized **kwargs from ormar.ForeignKey / ormar.ManyToMany
  to BaseField — fixes comment= (and other BaseField kwargs) being
  silently dropped on relation fields — thanks @booqoffsky #1239

🐛 Fixes

• Clearing a nullable ForeignKey with instance.fk = None now also clears the
  in-memory relation, so subsequent reads return None without a reload —
  originally @amit12297, rebased — see docs #1230
• Drop the unnecessary post-INSERT pk reload in save(), and raise
  ModelPersistenceError on backends that cannot return a generated pk
  (Oracle MySQL with non-AUTO_INCREMENT server-default pk) instead of
  silently assigning the row count to Model.pk — thanks @vnicolaichuk #919
• SelectAction aggregate identifiers now go through sqlalchemy.column()
  instead of string concatenation #1637
• Fix the mkdocs deployment workflow #1638