Changes:
    - Add required TensorFlow sources to SRC_URI needed for the
      Delegate application.
    - Pin FlatBuffers to an exact revision via flatbuffers.bbappend
      (defines the specific SHA needed to build TensorFlow).
    - Install the Delegate runner binary and shared library into the image.
    - Compile with -O2 to reduce the Delegate application size.
    - Add new recipe ethos-u-vela to build the Vela compiler and
      generate the test model file during the build based on
      configurable parameters.
    - The Delegate test model and IFM/OFM binaries are now produced at
      build time.

Signed-off-by: Michael Safwat <michael.safwat@arm.com>