Improve error message when `LightningCLI(run=False)` but user passes a stage as argument

Nobody has claimed this yet.

### Description & Motivation

This is a proposal to adjust the current error message

```
error: Unrecognized arguments: fit
```

to something more meaningful when `run=False` and passing a stage is not needed. Below is a minimal example that demonstrates this:

```py
from lightning.pytorch import LightningModule
from lightning.pytorch.cli import LightningCLI

class BoringModel(LightningModule):
pass

def main():
cli = LightningCLI(BoringModel, run=False)
cli.trainer.fit(cli.model)

if __name__ == "__main__":
main()

```

Run it like so:

```bash
python bug_report/bug_report_model.py fit
```

Error:
```
...
bug_report_model.py: error: Unrecognized arguments: fit
```

### Pitch

Adjust the error for this case to a message like this, pointing out that `run=False` was configured.

> Error: You passed 'fit' as stage but the LightningCLI is configured with `run=False`. Selecting a stage in this case is not supported.

### Alternatives

Keep as is

### Additional context

Raised by a question of a user in slack channel

cc @borda @carmocca @mauvilsa

Start with the LightningCLI entry point and reproduce the command shown in bug_report/bug_report_model.py using run=False and the fit argument. Done means the error explains that stage selection is unsupported when run=False, rather than only reporting an unrecognized argument.

Written by the indexing model from the issue text.