Harmonized error/logProbability in entire Factor/Conditional hierarchy

[dellaert]

Cleaning up before releasing 4.2 and implementing last hybrid PR

• renamed logDensity to logProbability, as density is continuous only
• we now have logProbability(x) = log(evaluate(x)) = K - error(x) for all conditionals
• probPrime(x) = exp(-error(x)) as always, but is only supposed to be used in factors

My advice for reviewing (seems daunting but most is rote):

• start with Factor.h, Conditional.h and the rest of inference - read the updated docs (and please flag issues).
• Then hybrid.
• Then linear, discrete, symbolic

Note that some methods are gone as now implemented in bottom base classes.

[ProfFan]

Most looks great to me!