We now check if both the target (filter above our own, OR parent if we're in first place) and the actual filter parent are valid instead of just the target.

Additionally, we get the source width and height for the filter, which if it's in first place will give the original source width and height (otherwise it would've picked the parent's and that will be with all filters *afterwards* applied, such as Crop or Aspect Ratio)

Also, I moved the known width and height check / update before the texrender logic.

Inside the render call we check if the target is the parent and if so, we render the filter via. `obs_source_skip_video_filter`, otherwise we just use `obs_source_video_render` on the target.

Offscreen isn't really necessary - sync also works for async sources.

This fixes an issue with e.g. Color Correction (and potentially other filters) being above the Dedicated NDI Output filter and rendering everything black.

Additionally, the `video_output_lock_frame` method was put in an inner `gs_stagesurface_map` call, as it returns a boolean on whether or not it was able to map it.

Performance seems to be on-par with the offscreen method, though with the stagesurface_map change I was getting slightly better performance (at least when I last tested it a few weeks ago for the other PR)

Since we can just use `get_width` and `height` on our own source (the filter) and get proper dimensions of either the filter above us or the parent source before any other filters are applied (first position) we'll just use that for simplicity and to keep things in line with the other code changes.