feat: add __len__ support to LightningDataModule

[dhruvildarji]

Closes #5965.

Adds __len__ support to LightningDataModule so users can call len(datamodule).

Default implementation: Attempts to return the length from train_dataloader().dataset. If the training dataloader is not configured or the dataset doesn't implement __len__, raises a TypeError with a helpful message directing users to override __len__ in their subclass.

Usage:

# Use the default implementation (returns train dataset length)
datamodule = MyDataModule()
datamodule.setup("fit")
print(len(datamodule))  # e.g., 50000

# Or override for custom behavior
class MyDataModule(LightningDataModule):
    def __len__(self) -> int:
        return len(self.train_dataset)

---

📚 Documentation preview 📚: https://pytorch-lightning--21546.org.readthedocs.build/en/21546/