optimize alphamerge for planar formats and avoid a memory copy

Fix double free in alphamerge_cuda

• Fix internal.h import on alphamerge_cuda and crop_npp

Fix yuva420p to yuv420p conversion in scale_npp

• Add alphamerge_cuda filter
• Add crop_npp filter

Adds a jpeg GPU accelerated encoder; nvJPEG.

Patches NVDEC to prevent issues with HW decoder surfaces

• Update to FFmpeg 7.1. Needed for pad_npp. Including all prior loom patches

Add pad_npp, a GPU accelerated video padding filter.

• Enable timeline editing for overlay_cuda