The TDA change used a binary search (worst case O(log n)) for finding
the right amount of RBs, whereas everywhere else we used a linear search
(O(n)). We generalize the binary search into a function nr_find_nb_rb()
and use it for DL and UL allocation.