Add a top-level Dockerfile to create a custom tfx image that includes tfx-addons

[cfezequiel]

The custom tfx image would include a dev version of tfx-addons for integration testing using Vertex AI Pipelines.

Example:

    runner = tfx.orchestration.experimental.KubeflowV2DagRunner(
        config=tfx.orchestration.experimental.KubeflowV2DagRunnerConfig(
            default_image=_CUSTOM_TFX_IMAGE),
        output_filename=pipeline.json)

This was suggested by @casassg in #230, where a similar Dockerfile exists to test the predictions-to-bigquery component. #230 (comment)