[Python] Generated type stubs don't work

Given a very simple FlatBuffer definition:

```fbs
table Foo {
bar: string;
baz: int64;
}

root_type Foo;
```

The generated type stubs (`flatc --python --python-typing foo.fbs`) include an incorrect alias to `uoffset`:

```py
from __future__ import annotations

import flatbuffers
import numpy as np

import typing

uoffset: typing.TypeAlias = flatbuffers.number_types.UOffsetTFlags.py_type
```

1. The corresponding `Foo.py` file does not define an `uoffset` alias.
```py
>>> from Foo import uoffset
Traceback (most recent call last):
File "", line 1, in
from Foo import uoffset
ImportError: cannot import name 'uoffset' from 'Foo' (/.../Foo.py)
```
2. `flatbuffers.number_types` is not accessible with a simple `import flatbuffers`, as [flatbuffers/\_\_init\_\_.py](https://github.com/google/flatbuffers/blob/master/python/flatbuffers/__init__.py) doesn't re-export `number_types`. An explicit `import flatbuffers.number_types` is needed.
3. The stub adds an unconditional `import numpy as np`, which breaks type checking when numpy is not installed.

Tested with Python 3.14.7, Pyright 1.1.411 and flatbuffers/flatc 25.12.19.

Start by reproducing the output from `flatc --python --python-typing` with the minimal `Foo` schema, then compare the generated stub with `Foo.py` and inspect `python/flatbuffers/__init__.py`. Check the generated imports and aliases under type checking without numpy installed; done means the stub matches the generated module and works with the stated dependencies.

Written by the indexing model from the issue text.