Optimize vec_transform

[Korijn]

• All operations are performed inplace on the vectors array
• Adds test for batch support (it already worked)
• Adds optional argument projection=True to vec_transform which, when set to False (to indicate no projection needs to be performed)
  • Skips homogeneous coordinate allocation
  • Performs matrix multiplication only on the 3x3 part of the matrix (scale+rotation)
  • Directly adds the translation component

@panxinmiao this should be much faster. you can use it in your sorting key function with projection=False