our_rand_r is a fast and simple random generator defined in sklearn/tree/_tree.pyx

Problems:

• The functions only return 0s if seeded with 0. We need to audit the use of this function and make sure it is never seeded with 0.
• The code is duplicated (sklearn/linear_model/cd_fast.pyx) and will be in [MRG+2] Adding Implementation of SAG - next episode #4738, in [WIP] Sparse PCA online in the direction of samples #4873, and possibly other PR . My first attempt to re-use only one code (by moving the inline function in _tree.pxd file) has failed: Every tests passed on Linux (Travis) but it broke 45 tests on Windows (AppVeyor), I did not figure out why.