My guess is people aren't willing to help because your code is quite long. I'm not sure what's wrong with it, kind of for the same reason, but I'll pitch in some suggestions:

1. Don't store the graph explicitly when you have a simple, straightforward implicit structure. Just navigate through all directions {{-1, 0}, {1, 0}, {0, -1}, {0, 1}} for each cell. Storing the graph only causes overhead, which can be significant if you use vector [] and there are many nodes, as is the case here.
2. Your BFS functions might not behave as expected. I TLE'd this problem during the contest because I didn't realize that in my implementation I had 2 types of edges, so basically I had written Bellman-Ford and its run-time went through the roof. This might also manifest by MLE, because the queue can get very large. See if there's any reason why that may happen.
3. Your code has a lot of casework which isn't necessary. General tip, try to simplify your code as much as possible before implementing it. It will make possible mistakes come out more easily.

For reference, here's my fixed submission, maybe it helps. http://codeforces.me/contest/590/submission/13855446