Filtering per batch and by percentiles

[pcm32]

Currently our filtering setup is very rough, it simply allows fixed thresholds (ie. >200 genes per cell). In reality, people want to filter out the topmost 5% or the lowest n% of a specific metric. It is also desirable to do this by batch / sample, instead of over the whole dataset, or even possibly limit the filtering to specific samples / obs fields.

• Implement % based filtering which will sort out elements based on a specific field and allow removal of > n% or < m % of that desired metric. This shouldn't replace of course the previous hard threshold setting and should be backward compatible (previous call should work as it is. This is of course work here and at Scanpy scripts / Seurat scripts.
• Implement the ability to enact the above (and any filtering command) on a per sample / batch / observation field. Again both here and on scanpy scripts / seurat scripts.