allow non-descriptor subtypes

### Description

consider the following example, where we want to define a `Protocol` for any object that has `system: str` and a `code: str` attributes:

```python
from typing import Protocol, runtime_checkable

from sqlalchemy.orm import DeclarativeBase, Mapped, MappedAsDataclass

class Base(MappedAsDataclass, DeclarativeBase):
pass

@runtime_checkable
class ConceptProto(Protocol):
@property
def system(self) -> str: ...

@property
def code(self) -> str: ...

class Concept(Base):
__tablename__ = "concept"

system: Mapped[str]
code: Mapped[str]

def fun(c: ConceptProto):
print(c.system, c.code)

fun(Concept(system="SNOMED", code="123456789"))
```

this currently fails with pyright:

```
#########################/proto.py
#########################/proto.py:30:5 - error: Argument of type "Concept" cannot be assigned to parameter "c" of type "ConceptProto" in function "fun"
  "Concept" is incompatible with protocol "ConceptProto"
    "system" is an incompatible type
      "Mapped[str]" is not assignable to "str"
    "code" is an incompatible type
      "Mapped[str]" is not assignable to "str" (reportArgumentType)
1 error, 0 warnings, 0 notes
```

it would be nice if sqlalchemy models could be used within that context, and manipulated by duck-typed libs without ever having to import sqlalchemy

you can see the initial discussion with the maintainer of sqlalchemy here: https://github.com/sqlalchemy/sqlalchemy/discussions/12889

Start by reproducing the supplied Python and SQLAlchemy example with pyright, then review the linked SQLAlchemy discussion. Trace how Protocol properties are checked against Mapped[str]; done means the example type-checks successfully while existing protocol diagnostics remain intact.

Written by the indexing model from the issue text.