Context windows fixes and features

[kijai]

Main features and bugfixes:

• Fixes channel dim cond_concat slicing for models such as Wan I2V, HunyuanVideo 1.5
• Adds option to use Freenoise
• Adds support for HuMo audio

Experimental features (disabled for now):

• Adds option to retain channel conds from first window, needed for models that always expect positional conditioning, ie. Wan I2V
• Adds option to split combined conds between multiple windows, enabling rudimentary prompt travel, and when combined with above option, multiple I2V input images as well.

[Kosinkadink]

Looks good, I only did minimal testing and I didn't test all the functionality across the diff models but since this is a feature that does not impact everyone and resolves a bunch of bugs/limitations, I will merge as is and we can get more feedback from users.