Refactoring: ContinuousNode introduced, moved stuff from Discrete and Continuous Node so generic Node, created MCMC example for multiple continuous Models

Contstructing a MarkovChain with MetropolisHastings is now working for Networks which consist only of LinearGaussian nodes or ProbabilityTables. Additionally it is now possible to specify the form of densityfunction that one expects for PosteriorMarginal. Learning a Gaussian from a MarkovChain is also possible

ExampleComparison is running correctly, Added MetropolisHastingsAlgorithm as StateTransition and began generalization of MCMC to continuous spaces

Changed behaviour of ProbabilityTable.to_jpt_by_states back to origninal behaviour, as is documented... I dont know who changed it

Extended examples/MCMC.py, improved MarkovChainSampler to use less memory (lukas idea), started working on division