5.2.20

[Base] added float overloads to RandomSample methods
[Base] changed array memory copies to use Buffer.MemoryCopy
[PixImage] respect channel order in format and type conversion
[PixImage] improved support for dual-channel images
[PixImage] fixed and improved format conversion
[Tensors] added simple accessors for Gray and BW
[Vrml] added support for gzip compressed files
[Vrml] allow registration of custom field parsers
[Vrml] improved robustness in case of uknonwn field tokens