`ModelCheckpoint` and `SaveConfigCallback` have different saving path.

Description & Motivation

I'm using LightningCLI and NeptuneLogger.

When I pass a NeptuneLogger object to Trainer class, I expect the checkpoints and config files to be saved in a path determined by the logger by default. However, the current behavior differs from my expectation.

I slightly investigated how these paths are determined. If a logger is given in a trainer, the ModelCheckpoint.dirpath is determined as shown in the code below.
https://github.com/Lightning-AI/lightning/blob/fe9e5d55bf7991ba36b76d6adae9075b93dfcaa0/src/pytorch_lightning/callbacks/model_checkpoint.py#L605

So I expect the config.yaml will be saved in os.path.join(save_dir, str(name), version). Basically, in NeptuneLogger, the version denotes an automatically assigned ID from Neptune.ai, so collaborating with version is very important.

However, the SaveConfigCallback acts as follows,
https://github.com/Lightning-AI/lightning/blob/6df43685ee2f2dd0c53eefec295e0d6c79796fd2/src/lightning/pytorch/cli.py#L243-L246
which is the same as just save_dir.

Pitch

config.yaml should be saved in os.path.join(save_dir, str(name), version).

Alternatives

None

Additional context

It is very complex to figure out the meaning, behavior, and connections of these paths, such as Trainer.default_root_dir, Trainer.log_dir, ModelCheckpoint.dirpath, 'Logger.log_dir, and Logger.save_dir. It would be nice for users if Lightning provided tutorials about it.

cc @lantiga @borda @mauvilsa

Start with the linked sections of src/pytorch_lightning/callbacks/model_checkpoint.py and src/lightning/pytorch/cli.py to compare how ModelCheckpoint.dirpath and SaveConfigCallback choose their paths. Trace the logger values used by NeptuneLogger, then verify that config.yaml follows the same save_dir/name/version location as checkpoints.