Advance optimization when using entire graph as frontier

[maawad]

Is your feature request related to a problem? Please describe.
There is extra work performed in the advance kernel when the input frontier is the entire graph

Describe the solution you'd like
The exclusive sum performed here is almost the same as the row offsets array when the input is the graph. We should just read the row offsets array (already in shared memory, sedges array) and subtract thread zero rows offset entry from all the values.

Additional context
From @neoblizz:

Yeah, when the graph is the input and already a CSR, there should be an if-condition to skip that for an optimization. That will help in push-based SpMV and PR.