Lock to numpy 1.23.5 to avoid removed np.bool in reinforcement_learning

The gym library used in the reinforcement_learning example uses np.bool
which is deprecated in numpy 1.20 and removed in numpy 1.24.  This is
further complicated by gym itself being deprecated and not accepting
updates that might fix the problem at its source; see
openai/gym#3258.

Fixed by locking the numpy version for this example to 1.23.5.

Author: charles-juicelabs

reinforcement_learning/requirements.txt

@@ -1,4 +1,4 @@
 torch
-numpy
+numpy==1.23.5
 gym
 pygame