Remove base_run_inference in TFX-BSL and use inference.base in Beam

TFX-BSL has shared code for a base run_inference implementation.

This is duplicate code and TFX-BSL should be able to use Beams run_inference.