Improved R support (object serialization, environment isolation)

[D3SL]

If I understand correctly R blocks work by RScript on their contents and serializing any input or output as a CSV file. If so it seems like each R block can run completely arbitrary code, natively at that, but at the cost of complete isolation and only being able to ingest or export a single tabular object per block.

This means for example it's not possible to set a variable either in the R global environment or Mage itself for use in other blocks, output non-dataframe objects, or output multiple objects (even bundled via list).

There are multiple ways to address this I think. The most comprehensive is using rpy2 combined with reticulate for an actual two-way connection between R and Mage's internal python environments. R's QS package, which allows for serializing pretty much every type of R object, may also be helpful.

Parallel to this, for dataframes in particular the use of Apache's Arrow format would be a significant optimization over CSV. There are packages in both R and Python for it.