The PW buffer is already processed before the grd_egl_thread_download()
call, so move the acquire_pipewire_buffer_lock() call further up.
This also makes this behaviour consistent with the MemFd path, where the
acquire_pipewire_buffer_lock() call also happens before any processing
action is done.

This makes the overall code more readable, as callback here always
refers to the callback, when a frame is ready.

on_frame_ready() releases the PipeWire buffer, if it was acquired
previously. So, no need to also try to release the buffer before calling
on_frame_ready().