conditionally enabling dataflow for ImportExampleGen

[deep-diver]

Closes #20

This PR includes the following changes:

• by scanning a environment variable FULL_RES_DATA
  • it will choose the GCS bucket where the full resolution of images are
  • it will choose different training function in model_full_res.py
  • it will choose dataflow specific configurations, then insert it to ImportExampleGen with with_beam_pipeline_args method

the FULL_RES_DATA environment variable is controlled by GitHub Action. You can provide the value true or false within GitHub Action panel.

[deep-diver]

@sayakpaul

addressed your comments! thanks for the careful inspection :)