Fixed false negatives around edges and added second phase of reporting

[RunDevelopment]

Fixes #479.

This PR includes 2 changes:

1. I improved the handling of edges. Previously, all verify function would just immediately return without reporting anything if the next character included the start/end (edge) of the input string. I likely did this because the logic around string edges is a bit more complex.
2. In order to fully report all useless assertion in the examples here, I had to add assertion reordering. E.g. to detect that the \b in /\b^a/ is useless, \b has to know that there is a ^ after it. The solution here is to (conceptually) reorder the assertions to /^\ba/ which allows \b to "see" ^.
   However, assertion reordering also causes overreporting. E.g. Since both $s in /$$/ can see each other, both will be reported even though only one is actually useless. To fix this, reporting is done in 2 phases. In the first phase, we do everything just like we did before. In the second phase, we reorder while being careful not to reorder already reported assertions.

[ota-meshi]

Awesome! Thank you!