SP adaptSynapses: Experiment with discouraging shared inputs

[subutai]

Clone of (now closed) NuPIC issue https://github.com/numenta/nupic/issues/894

As per https://github.com/numenta/nupic/issues/892, might be worth experimenting with the following approach:

"Permanence values for shared inputs, which are input bits that are turned on and connected to more than one active column, are decreased in order to discourage multiple columns from learning the same input pattern."

Subutai: "For reference it was implemented in FDRCSPatial2.py as the parameter synPermActiveSharedDec. I don't know that we ever proved it was useful."