Restart training with new data, mid-epoch

## 🚀 Feature

Right now, you can change the datasets used in a `CombinedStreamingDataset` and resume training on epoch boundaries. It would be great if you could resume training with new datasets mid-epoch.

### Motivation

When we're doing curriculum learning, we don't know the right number of steps or epochs to train. If we reach a sufficient validation loss, we kill the training and resume training with a new group of datasets (i.e. adjust the curriculum). Accordingly, we often have to kill training mid-epoch and restart with new datasets.

### Pitch

If you are training with N datasets and kill training K steps into epoch N, change the underlying datasets, and resume training from a checkpoint that was saved mid-epoch, the trainer should jump to epoch N + 1 with the new datasets, the old optimizer state, the correct global batch index.

### Alternatives

Right now, we have two workarounds:
- **Delete the `loops` part of the last saved mid-epoch checkpoint, before we resume training with different datasets.** This isn't a great solution - because you resume training at epoch 0, which messes with any learning rate schedulers we have.
- **Copy the `loops` part of the last saved epoch checkpoint into the more recent epoch that was saved mid-epoch, before we resume training with different datasets**. This approximately mitigates the learning rate scheduler issue -- but isn't the cleanest solution and is a pain to do manually every time.

### Additional context

Start with the CombinedStreamingDataset behavior and the trainer's checkpoint-resume path described in the issue. Trace how mid-epoch checkpoints store loops, optimizer state, epoch, and global batch index; done means replacing datasets resumes at the next epoch with the old optimizer state and correct global batch index.

Written by the indexing model from the issue text.