RF: Convert nan values in bvectors to 0's

This get's rid of warnings of the following type that currently show up
in our tests:

    ./Users/arokem/source/dipy/dipy/core/gradients.py:132: RuntimeWarning: invalid value encountered in less_equal
  bvecs_close_to_1 = abs(vector_norm(bvecs) - 1) <= atol

dipy/core/gradients.py

@@ -129,6 +129,7 @@ def gradient_table_from_bvals_bvecs(bvals, bvecs, b0_threshold=0, atol=1e-2,
                          "respectively, where N is the number of diffusion "
                          "gradients")
 
+    bvecs[np.isnan(bvecs)] = 0
     bvecs_close_to_1 = abs(vector_norm(bvecs) - 1) <= atol
     if bvecs.shape[1] != 3 or not np.all(bvecs_close_to_1[dwi_mask]):
         raise ValueError("bvecs should be (N, 3), a set of N unit vectors")