Add checks (and converters?) for documents with multiple sentences in debug-data

## Feature description

The parser section of `spacy debug-data` should show a warning when there are no/few documents with multiple sentences in the training data.

Potentially add a simple converter to `spacy convert` to group sentences, similar to `-n` with the IOB converters. A bit of variety in document lengths is probably a good idea here, too, rather than just `-n N`, but I don't know if it makes that much difference in the model performance.