Implement transforms for augmentation

[gutzcha]

Implement the following transforms for data augmentations

Data shape:
Batch size X Number of frames X

• Horizontal mirror - flip each pose frame left-right
• Spatial normalization - normalize coordinates to the range [0,1]
• Temporal mirror - flip order of frames to reverse the pose sequence
• Tilt pose - tilt head position in each frame by x degrees

Feel free to suggest other transforms