generalise Dirichlet Distribution to deal with vectors that do not sum to 1

[EvaLiyt]

Solution: normalise the input vector so that its components sum to 1.
Compensate for the space reduction by adding a constant log(sum)* (dimension-1).

[rbouckaert]

Fixed by de9772d

[rbouckaert]

On reflection, this should be made more robust by declaring the expected sum of values.
Since now things are normalised, the distribution does not recognise values with zero probability any more.

Proposed solution: add an new Input that defaults to 1 and make sure the sum of values is (almost) equal to that expected sum.

[walterxie]

1. The sample method did not consider expectedSum
2. expectedSum requires "protected" for WeightedDirichlet