exponential slowdown when using `yield from` with type alias

**Describe the bug**

if multiple function use `yield from` with a type alias type, processing time grows almost exponentially with the number of such functions.

**Code or Screenshots**

```python
#!/usr/bin/env python

from collections.abc import Iterator

class A[T]:
def __iter__(self) -> Iterator[T]: ...

class B[T]:
def __iter__(self) -> Iterator[T]: ...

type C[T] = A[T] | B[T]

def g() -> C[int | str]: ...

def f_0(): yield from g()
def f_1(): yield from g()
def f_2(): yield from g()
def f_3(): yield from g()
def f_4(): yield from g()
def f_5(): yield from g()
def f_6(): yield from g()
def f_7(): yield from g()
def f_8(): yield from g()
def f_9(): yield from g()
```

```
$ time pyright po.py
0 errors, 0 warnings, 0 informations

real 25m25.431s
user 25m24.261s
sys 0m3.227s
```

| # of `f_N` | seconds |
|--:|--:|
| 5 | 8 |
| 6 | 29 |
| 7 | 98 |
| 8 | 278 |
| 9 | 711 |
| 10 | 1524 |

if `g` is declared as `def g() -> A[int | str] | B[int | str]: ...`, the problem does not occur.

**VS Code extension or command-line**

command-line tool, 1.1.407

Start by running the command-line tool against the reported po.py-style snippet and compare the type-alias return annotation with the direct union annotation. Trace the analysis of yield from and generic type aliases; done means the alias form no longer shows exponential slowdown while producing the same diagnostics.

Written by the indexing model from the issue text.