Release 1.0.0b23

Visible changes:

• sum() now returns 0 for an empty vector instead of null
• A new collating mean() function with similar behaviour to sum()
• Resolve inconsistencies between use of minimum/min and maximum/max attributes in different places by making both versions acceptable everywhere
• Switch to using the model matrix cofactor for normal matrices instead of using the transposed inverse. This is faster and more correct – in particular, it means that scaling by a negative value will result in the model normals being reversed to correctly match the reversed face winding
• New syntactic sugar for nested for loops: for x in a, y in b
• The grammar rules have also been changed slightly to allow ; sequences to be used as for loop sources without needing to be wrapped in brackets
• New nearest=false attribute on window nodes can be used to turn off linear interpolation on sampling
• A load of documentation fixes and improvements

Full Changelog: v1.0.0b22...v1.0.0b23