adjacent_filtered's postcondition doesn't correctly describe behaviour

[tonyelewis]

I think the documentation for adjacent_filtered's postcondition, ie:

For all adjacent elements [x,y] in the returned range, bi_pred(x,y) is true.

…isn't quite right. For example, this:

#include <boost/range/adaptor/adjacent_filtered.hpp>

#include <iostream>
#include <vector>

int main() {
   const std::vector<int> a = { 0, 1, 2, 3, 4, 5, 6 };

   const auto b = a | boost::adaptors::adjacent_filtered(
      [] (const int &x, const int &y) {
         return ( y % 2 == 1 ) && ( y == x + 1 );
      }
   );
   for (const auto &x : b) {
      std::cerr << x << "\n";
   }
}

…outputs:

0
1
3
5

Yet two of the pairs of adjacent elements [x,y] in this range fail bi_pred(x,y) (because they differ by 2, not 1).

I think it's more like: bi_pred is true on each element in the returned range preceded by the element that preceded it in the original range (not in the returned range).

[tonyelewis]

This is related to #80.