Typo in LandmarkLBF.cpp

[TonyCongqianWang]

In the supportVectorRegression(...) function of LandmarkLBF.cpp there is a typo. lambda and upper_bound were declared as a size 1 array, but it was supposed to be a size l array! This can be verified by looking at the "original" code the algorithms was based of. And indeed, on github, it is almost impossible to tell the difference between l and 1. This typo results in some nan values and occasional crashes due to segfaults (not suprising, considering the array size is too small!).

scikit-learn/scikit-learn@2b60d32#diff-bf60bef97c003384f56a1fd92fe102f6ac1eea40298e911dd95b4d28601bbc14
(line 830 in linear.cpp)

[asmorkalov]

@TonyCongqianWang Thanks a lot! Feel free to submit PR.