support newer versions of tensorflow (2.1+) (#3830)

* support tf2.x and python3.8

* tensorflow==2.2.0rc3 for python3.8

* stick with tf2.1 and py3.7 for now

* More gail visual steps in simple test (#3836)

* increase gail visual ppo steps

* increase to 2000

* tune steps down to 750

Co-authored-by: andrewcoh <54679309+andrewcoh@users.noreply.github.com>

.circleci/config.yml

   python373:
     docker:
       - image: circleci/python:3.7.3
+  python382:
+    docker:
+      - image: circleci/python:3.8.2
 
 jobs:
   build_python:

ml-agents/mlagents/trainers/tests/test_simple_rl.py

         step_size=0.2,
     )
     override_vals = {
-        "max_steps": 500,
+        "max_steps": 750,
         "learning_rate": 3.0e-4,
         "behavioral_cloning": {"demo_path": demo_path, "strength": 1.0, "steps": 1000},
         "reward_signals": {

ml-agents/setup.py

         "License :: OSI Approved :: Apache Software License",
         "Programming Language :: Python :: 3.6",
         "Programming Language :: Python :: 3.7",
+        "Programming Language :: Python :: 3.8",
     ],
     # find_namespace_packages will recurse through the directories and find all the packages
     packages=find_packages(exclude=["*.tests", "*.tests.*", "tests.*", "tests"]),
         "Pillow>=4.2.1",
         "protobuf>=3.6",
         "pyyaml",
-        "tensorflow>=1.7,<2.1",
+        "tensorflow>=1.7,<3.0",
         'pypiwin32==223;platform_system=="Windows"',
     ],
     python_requires=">=3.6.1",

test_constraints_max_tf2_version.txt

 # For projects with upper bounds, we should periodically update this list to the latest release version
 grpcio>=1.23.0
 numpy>=1.17.2
-tensorflow>=2.0.0,<2.1.0
+tensorflow>=2.1.0
 h5py>=2.10.0