Add _log_api_usage_once() calls on remaining endpoints

[datumbox]

🚀 The feature

Add _log_api_usage_once() calls to:

• The constructors of all Segmentation model constructors. The implementation should be identical to what was done for the other models. See Expand usage logging to models #4735
• All the Transforms. For those that have functional equivalent the call should be placed directly in their method. For those that don't have a functional equivalent (for example Compose), the call should be placed in the class constructor. The approach is identical to what was done for the Operators. See Add logging to torchvision ops #4799
• To the public python IO methods at torchvision.io.image and torchvision.io.video and to the VideoReader class constructor. On the python side these are the following endpoints: 1, 2, 3. It might be also worth covering C++ endpoints that might be called directly (without Python) such as 4, 5, 6.

cc @vfdev-5 @pmeier