Mean statistics fixes

[Jolanrensen]

Helps and follows #961

To be merged after #1078

• Redid implementation of mean. It's now based on TwoStepNumbersAggregator, such that mixed numbers are unified first.
• Big number support is dropped.
• Created generic rowX() function aggregateOfRow(), used by rowMean() and rowMeanOf<T>()
• Removed big numbers from describe()
• Mean statistics fixes #1091 (comment)

[AndreiKingsley]

Please add some notes in comments for public mean about new behavior, so that we don't forget to mention it there when writing KDocs in the future.

[Jolanrensen]

Fixed feedback, but I also found some other important stuff:

• We still had a lot of internal mean functions that we didn't use anymore, because we calculate mean as double.
• Aggregators had a mandatory nullable return type, however this adds unnecessary logic for aggregators that never return null, such as mean, so I rewrote some of that.
• Also, aggregators were allowed nulls for iterables, but for columns they were filtered out. They are now filtered out in all cases.
• More will follow in later aggregator/statistics-PRs

[AndreiKingsley]

Great! However, I wonder - are there any aggregators that take null into account?

[Jolanrensen]

Great! However, I wonder - are there any aggregators that take null into account?

They all filter nulls out before aggregating. This was always done already for columns and it's now always the case.

There doesn't seem to be a statistic function that does something special with nulls, so I think this is the best choice.