Thanks a lot for your feedback, these are definitely things we're looking to tackle.

We're planning on improving scrolling performance soon, it's definitely a priority for us to improve our performance for large datasets.

As for the linking, we definitely want to do more to help you normalize your data. We can also infer column types to help with the import process. Doing this type of thing on paste is an interesting idea.

We have some ideas for how to make it so that you don't need a primary column. One thing that you can do now is make the primary column a formula, and then reference other columns to generate a key.