Create test for SP boosting in over-provisioned setting

[breznak]

https://discourse.numenta.org/t/how-boosting-changes-the-sp-representation/1159/10?u=breznak

Problem

Boosting creates noice when there is too much columns and too simple input patterns. So there will always be some columns "bored" and weak.

Test scenario

SP 2048 cols
Desire 40 active columns (2% sparsity)
10 input patterns (10x random SDR)

After learning, columns will estabilish, and some columns will always be unused, and get boosted, and create fake anomaly.

Proposals

• do not include boosted columns into anomaly (but artificial solution)
• limit number of columns that get boosted to a small fraction of active columns, so boosting anomalies do not have effect on AD
• fix boosting mechanism not to do this
• remove boosting and replacce with synaptic death Replace Boosting with cell death (energy)  #288