Easier access to train_batch_idx for control

Nobody has claimed this yet.

### Description & Motivation

Global step is defined in terms of optimizer steps, for multi optimizer training most flow control is easier defined in terms of train batches

### Pitch

There are possibilities to access the train batch index for example with `self.trainer.fit_loop.batch_idx` but for example there is no option to use instead of Trainer `max_steps`. The suggestion is to provide an option to use train_batch_idx instead of steps in all APIs.

### Alternatives

Leave unchanged

### Additional context

_No response_

cc @borda

Start with the exposed self.trainer.fit_loop.batch_idx entry point and the Trainer max_steps API mentioned in the issue. Trace how step limits are consumed across the training APIs, then define the affected scope and verify that train-batch-based control works consistently with multi-optimizer training.

Written by the indexing model from the issue text.