Strengthen Validation Of Gossip Scoring Parameters

[nisdas]

During further testing of our parameters for peer scoring, we came across some unusual scores for a significant amount of our peers, with a large amount giving their score as NaN. On further investigation it came about that when initializing the scoring parameter there had been an arithmetic error when determining the parameter. This lead to the parameter being initialized with a negative infinite value. The parameter validation was unable to capture this, which is the main reason this PR has been opened up.

• [x] Add in validation for both NaN and Infinite floating point values and throw an error if any of the topic parameters are initialized as such.
• [x] Add regression test for it.

[vyzo]

@nisdas do you mind rebasing on master? I just merged #406 which disables the flaky test that keeps failing on travis.

[vyzo]

thank you!