Right now Clients of a VDA are unable to do anything with the provided Pictures if they do not benefit from some kind of hardware acceleration. In particular, it is impossible to map the output buffers for the CPU.

While we certainly don't want to use the CPU in the regular video pipeline, having the possibility to map buffers would be interesting:

* We could dump frames for debugging purposes,
* We could completely remove the GPU dependency in the VDA unittest while keeping the ability to validate thumbnails.