major changes for ralg

I have committed some major changes for ralg (+ some more code for prob.point, to simplify my further OO development).

Since I have been busy with connecting ipopt and these ralg changes, patterns are intended to be implemented in next OO release.

Below is output of OO v0.17 ralg vs current ralg implementation (made by /examples/nlp_bench2.py for N=100). Usually OO v0.17 ralg consumes about 5-10% objFunc and 3-5% nonlinear constraints evaluations greater.