99cc417 prevented flow from an output argument from flowing into an argument to the same call. This revealed a problem that only became apparent on the QA run for 2.14.4. When we have a snippet such as:

void test(int* p) {
  write_to_first_argument_and_read_from_second_argument(
    p, // (1)
    p // (2)
  );
  sink(p); // (3)
}

we can't just block flow from a post-update node of a call (i.e., (1)) to a use in the same
call (i.e., (2)) since the use-use chain is (1) -> (2) -> (3).

So instead we have to skip the step from (1) to (2) and go directly from (1) to (3). This PR does this using a call to shortestDistances.

(Leaving this as a draft while DCA is running.)