combined_loader: concatenating inputs

Nobody has claimed this yet.

### Description & Motivation

Application: using the combined data loader with two datasets of the same format
Suggestion: option for concatenating the data + labels in the combined loader
Motivation: currently one has to edit model code when data source is changed -> very unelegant solution and likely leads to errors!
Comment: Maybe I just did not understand the documentation right. if so please improve the documentation. :-)

currently I use:

```
iterables = {"dataset1": dataloader1, "dataset2": dataloader2}
combined_loader = CombinedLoader(iterables, mode="min_size")

```
and have to unpack the data in the models training step using:

```
def training_step(self,batch,batch_idx):
#batch1,batch2 = batch
x = torch.concat([batch['dataset1'][0],batch['dataset2'][0]])
y = torch.concat([batch['dataset1'][1],batch['dataset2'][1]])

```

### Pitch

It would be great to have something like this (example for regression tasks on images):
```
dataloader1 = DataLoader(..., batch_size = b1)
dataloader2 = DataLoader(..., batch_size = b2)

iterables = {"dataset1": dataloader1, "dataset2": dataloader2}
combined_loader = CombinedLoader(iterables, mode = 'min_size', operation = 'concat')
x,y = next(iter(combined_loader)
# x.shape = [b1+b2,h,w,c]
# y.shape = [b1+b2]

```

### Alternatives

_No response_

### Additional context

_No response_

cc @borda @justusschock @awaelchli

Start with the CombinedLoader API and its current min_size behavior, then compare it with the shown training_step concatenation and proposed next(iter(combined_loader)) result. Done means the requested operation is clearly defined and the combined loader can produce concatenated data and labels for the two same-format datasets without model-side unpacking.

Written by the indexing model from the issue text.