setup venv in yamato commands

ml-agents/tests/yamato/training_int_tests.py

     else:
         standalone_player_path = "testPlayer"
 
-    venv_path = init_venv(python_version)
+    init_venv(python_version)
 
     # Copy the default training config but override the max_steps parameter,
     # and reduce the batch_size and buffer_size enough to ensure an update step happens.
         f"mlagents-learn {yaml_out} --force --env={env_path} "
         f"--run-id={run_id} --no-graphics --env-args -logFile -"
     )  # noqa
-    res = subprocess.run(
-        f"source {venv_path}/bin/activate; {mla_learn_cmd}", shell=True
-    )
+    res = subprocess.run(f"{mla_learn_cmd}", shell=True)
 
     # Save models as artifacts (only if we're using latest python and C#)
     if csharp_version is None and python_version is None:

ml-agents/tests/yamato/yamato_utils.py

         pip_commands += ["-e ./ml-agents-envs", "-e ./ml-agents", "-e ./gym-unity"]
     if extra_packages:
         pip_commands += extra_packages
+
+        print(f'Running "python3 -m pip install -q {cmd} {pip_index_url}"')
-            f"source {venv_path}/bin/activate; python3 -m pip install -q {cmd} {pip_index_url}",
-            shell=True,
+            f"python3 -m pip install -q {cmd} {pip_index_url}", shell=True
         )
     return venv_path