Using type alias causes error where directly using the type doesn't

It is surprising (at least to me) that in the following pkl definition:

```pkl
typealias Step = GetStep | PutStep

class Job {
steps: Listing?
}

class GetStep {
prop1: Step?
}

class PutStep {
prop1: Step?
}
```

An error is thrown for prop1 on the PutStep saying:
```
–– Pkl Error ––
Type alias definitions must not be cyclic.

8 | prop1: Step?
^^^^
at foo#GetStep (file:///Users/jonathan.harden/test/pkl/foo.pkl, line 8)

1 | typealias Step = GetStep | PutStep
^^^^^^^
at foo#Step (file:///Users/jonathan.harden/test/pkl/foo.pkl, line 1)

4 | steps: Listing?
^^^^
at foo#Job (file:///Users/jonathan.harden/test/pkl/foo.pkl, line 4)

106 | text = renderer.renderDocument(value)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
at pkl.base#Module.output.text (https://github.com/apple/pkl/blob/0.25.2/stdlib/base.pkl#L106)
```

However if instead of the typealias, I use the value of the typealias directly it is fine:

```pkl
class Job {
steps: Listing?
}

class GetStep {
prop1: (GetStep | PutStep)?
}

class PutStep {
prop1: (GetStep | PutStep)?
}
```

It seems I should be able to use a typealias anywhere I would use a type and get the same result.

I can't see in the documentation for type alias where this is mentioned.

Clearly the workaround is to not use a type alias and instead just redefine the complex type everywhere which is ok for now, but it's a surprising limitation.

Versions:

```
$ pkl --version
Pkl 0.25.2 (macOS 14.1, native)
```

Reproduce the example with Pkl 0.25.2, comparing the typealias form with the direct union form. Start by tracing the typealias cycle diagnostic and the rendering path mentioned in stdlib/base.pkl, then confirm that both forms behave equivalently without the cyclic-alias error. There is no named test file, so add or run the relevant regression coverage if the project identifies one.

Written by the indexing model from the issue text.