Profiling script for AlphaZero agent and updates benchmark

[adamantivm]

Unfortunately the profiling results don't seem very useful:

Total time: 56.34 seconds
         4499384 function calls (4420412 primitive calls) in 58.276 seconds

   Ordered by: internal time

   ncalls  tottime  percall  cumtime  percall filename:lineno(function)
     3978   49.814    0.013   49.814    0.013 {built-in method tensorflow.python.client._pywrap_tf_session.TF_SessionRun_wrapper}
     3978    3.085    0.001   53.114    0.013 session.py:1359(_do_run)
    15907    0.367    0.000    5.817    0.000 session.py:1381(<genexpr>)
    24782    0.301    0.000    0.301    0.000 {built-in method numpy.asanyarray}
   257789    0.201    0.000    0.266    0.000 mcts.py:120(uct_value)
    23859    0.200    0.000    0.533    0.000 _version.py:55(__init__)
   469043    0.173    0.000    0.259    0.000 {built-in method builtins.isinstance}
        1    0.153    0.153    0.153    0.153 {built-in method tensorflow.python._pywrap_tfe.TF_ListPhysicalDevices}
        1    0.133    0.133    0.133    0.133 {built-in method tensorflow.python.client._pywrap_tf_session.TF_NewSessionRef}
    41592    0.127    0.000    0.127    0.000 mcts.py:111(__init__)
   257789    0.124    0.000    0.390    0.000 mcts.py:340(<lambda>)
    25279    0.121    0.000    0.512    0.000 {built-in method builtins.max}
    23446    0.119    0.000    0.555    0.000 _shape_base_impl.py:518(expand_dims)
    20367    0.102    0.000    0.102    0.000 {built-in method numpy.array}
...