Enable mp4 input in video pipeline

- Enable HW video decoder for video file src
- Adjust the render_size scaling mechanism
- Take care of stride in the inference frame
  (glbox generates stride as multiples of 4)

Currently we need to pass the decoded stream through mtkmdp, and
this plugin has many bugs (not supporting GL, format error to
rsvgoverlay,etc.). The POR is to use v4l2convert instead.

Bug: 151393184
Tested: edgetpu_demo --device
Tested: edgetpu_detect (with webcam)
Tested: edgetpu_classify (with image)
Change-Id: Ic448069c1a950e077a4ff62ba84aedffebbc9c31