`top_k` parameter of `ModelCheckpoint` default value

Nobody has claimed this yet.

### Description & Motivation

I believe that the `top_k` would highly benefit from the improvement in its default value. Currently, it defaults to `1`. However, sometimes, like for example when I am trying to save models every n train steps, it doesn't make sense to save the `top_k=1` model. I would presumably like to have it save all the models.

### Pitch

So, I would suggest that the default value should be `None`. In case where `every_n_train_steps` or `every_n_epochs` or `train_time_interval` has some value with `monitor=None` and the `top_k` value is not given, i.e., still bears `None` it should change to -1. However in cases where the `monitor` parameter is set the `top_k` value should be changed to 1, if not specified.

### Alternatives

_No response_

### Additional context

_No response_

cc @lantiga @borda

Start at the ModelCheckpoint entry point and trace how top_k interacts with monitor, every_n_train_steps, every_n_epochs, and train_time_interval. Clarify the default behavior for monitored versus unmonitored checkpoints, then verify that the resulting behavior matches the requested cases and does not regress existing checkpoint selection.

Written by the indexing model from the issue text.