fastplyr 0.9.9

Improvements

• Optimisations for group-unaware functions

Simple functions such as base R math operators +, /, abs, etc, are
now internally marked as group-unaware. This has a very significant speed
improvement for large grouped data frames.

This means that expressions containing only group-unaware functions, e.g.
(x + y) / abs(z), are evaluated on the entire data frame instead
of on a by-group basis.

If the expression contains any functions not marked as group-unaware, e.g.
x + cumsum(y) (as cumsum() is not flagged as group-unaware),
then usual evaluation applies except in the case of other statistical functions
which are optimised in a separate way.

• Exported common dplyr helpers such as across, pick, etc.

Bug fixes

• Accessing columns through .data should work correctly now.

• f_reframe would not recycle correctly in some cases and has now been fixed.

• An issue where f_arrange would add variables has been fixed.

• An issue where across was selecting grouped variables has been fixed.

• Fixed an issue where in some cases lists where not being handled correctly in
  calls to across().