Graceful stop on an extrenal file marker

Nobody has claimed this yet.

### Description & Motivation

Sometimes the training settings are hard to set properly, for example the `patience` parameter is set to the big value, and/or `min_delta` parameter is too small.

So I understand that the training should be stopped but there is no way for me to gracefully stop it myself.

### Pitch

Would nice to have an option to stop training with some kind of signal, the easiest way is just to create some file with `touch stop.txt`

```python
class ForceStopCallback(pl.Callback):
"""Gracefully stops training when a flag file appears on disk.

Checks for the flag file at the end of each validation epoch and sets
``trainer.should_stop = True`` if the file exists, allowing the current
epoch to finish cleanly before stopping.
"""

def __init__(self, flag_file_path: pathlib.Path):
self.flag_file_path = flag_file_path

def on_validation_end(self, trainer: pl.Trainer, module: pl.LightningModule) -> None:
if self.flag_file_path.is_file():
msg: str = f'stopped by the explicit request from the user: {self.flag_file_path}'
lightning.pytorch.utilities.rank_zero_warn(msg, category=RuntimeWarning)

trainer.should_stop = True
```

and later

```python
self.force_stopping_callback: t.Final[ForceStopCallback] = ForceStopCallback(
flag_file_path=self.snapshot_folder_path / 'stop.txt',
)
```

### Alternatives

_No response_

### Additional context

_No response_

cc @lantiga

Start by tracing the callback hook around on_validation_end and the handling of trainer.should_stop, then inspect how snapshot_folder_path is used for the stop.txt marker. Done means an external flag file can request a clean stop after validation, with the proposed warning behavior covered by tests.

Written by the indexing model from the issue text.