Allow to specify the value of the `trainer/global_step` metric on W&B during validation

Nobody has claimed this yet.

## 🚀 Feature

Allow to specify the value of the `trainer/global_step` metric on W&B even during validation, as it is possible during training.

### Motivation

The default value for the `trainer/global_step` metric during validation is defined as the "total validation step":

https://github.com/Lightning-AI/lightning/blob/e7ad151c311890c931aa5ff3df3672fd21fd351c/src/pytorch_lightning/loops/epoch/evaluation_epoch_loop.py#L152

whereas during training the default value is `trainer.fit_loop.epoch_loop._batches_that_stepped`, even if there are some inconsistencies (see #13752).

The fact that the behaviour of `trainer/global_step` is different when switching from training to validation, generates graphs like this on W&B:

as well as the fact that W&B complains when trying to use `trainer/global_step` as `x` axis because it's not monotonically increasing:

For training it's possible to set a custom value for `training/global_step` by using `self.log("step", ...)` during the training step. For validation, there's no such possibility.

I understand that if one logs using `on_step=True` during validation, it's important to keep track of the different validation steps. But if one logs using `on_step=False` and `on_epoch=True`, it would be useful to link the val metric to the current training step instead of the total validation step.

### Pitch

One can set a custom value for `training/global_step` even during validation.

### Alternatives

Another possibility could be to log a different metric for validation, e.g. `trainer/validation_global_step` and to use that as default x axis as it is done with `trainer/global_step` here:

https://github.com/Lightning-AI/lightning/blob/6df6deac8fdf0c44b509d531441b4f7e446de6b4/src/pytorch_lightning/loggers/wandb.py#L396-L397

cc @awaelchli @morganmcg1 @borisdayma @scottire @parambharat @manangoel99

Start by comparing validation step handling in src/pytorch_lightning/loops/epoch/evaluation_epoch_loop.py with training's fit_loop.epoch_loop._batches_that_stepped, then inspect the W&B logger entry point in src/pytorch_lightning/loggers/wandb.py. Done means validation logging can use a specified global-step value without breaking per-step validation tracking or W&B's monotonic-axis expectations.

Written by the indexing model from the issue text.