feat: nulls_ordering (NULLS FIRST/LAST) for order_by (#769)

[collerek]

Summary

Ports #769 to the current codebase and incorporates the review feedback that was never addressed upstream. Adds Model.field.asc(nulls_ordering=...) / .desc(nulls_ordering=...) backed by a new ormar.NullsOrdering enum (FIRST / LAST).

• NullsOrdering(str, Enum) in ormar/queryset/clause.py, re-exported from ormar.queryset and top-level ormar.
• OrderAction gets an optional nulls_ordering; a new _build_order_expression() helper assembles the final SQL.
• Emits standard NULLS FIRST / NULLS LAST on PostgreSQL and SQLite (3.30+). On MySQL, emulates via col IS [NOT] NULL, col <dir> — a pure sort-key prefix, no effect on the result set.
• Fixes the nested-prefix bug the reviewer flagged on the original PR: the fully-qualified "table"."col" is threaded through both halves of the MySQL expression, so joined/aliased tables keep their prefix.
• FieldAccessor.asc() / .desc() accept Optional[NullsOrdering] and raise ValueError for other types.
• QuerySet.order_by() / QuerysetProxy.order_by() needed no signature change — they already accept OrderAction instances, so Model.field.asc(nulls_ordering=...) propagates through both.
• Docs: new subsection in docs/queries/filter-and-sort.md with usage and the MySQL dialect note.

Test plan

• pytest tests/test_queries/test_order_by.py — 5 tests pass on SQLite
• make pre-commit — ruff format, ruff check, mypy strict all clean
• make coverage — 503 passed, 1 skipped, 100% total coverage
• CI make test_mysql — exercises the MySQL emulation branch (marked # pragma: no cover in the default SQLite run)
• CI make test_pg — exercises the native NULLS FIRST/LAST path on PostgreSQL

New tests added:

• test_sort_order_nulls_first_last — dedicated main-model test with unique sort_order values covering all four {asc, desc} × {FIRST, LAST} combinations plus two ValueError guards.
• test_sort_order_on_related_model — extended with the same four combinations using Toy.owner.age, exercising the joined-table prefix path that the original PR's MySQL branch broke.
• Every test in test_order_by.py is now wrapped in database.transaction(force_rollback=True) so writes roll back between tests.

[collerek]

Superseded — force-pushed the same work onto #769 so the original PR's discussion thread stays in one place.