Cycles are reported incorrectly

[byorgey]

Describe the bug
I just noticed that Swarm.Util.Graph.getGraphCycles does not actually find directed cycles, and thus failOnCyclicGraph could output an error message indicating a cycle that does not exist. Note that whether an error is generated is correct --- failOnCyclicGraph indeed fails precisely when the graph has directed cycles. However, when there are directed cycles, it may print something which is not a directed cycle.

To Reproduce
The smallest example I could come up with is the following scenario description:

version: 1
name: "bad-cycles"
author: Brent Yorgey
description: An example for 2198
robots:
  - name: base
objectives:
  - id: a
    condition: 'true'
    prerequisite:
      logic:
        and:
          - b
          - c
          - d
  - id: b
    condition: 'true'
    prerequisite: a
  - id: c
    condition: 'true'
    prerequisite:
      logic:
        and:
          - a
          - d
  - id: d
    condition: 'true'
    prerequisite:
      logic:
        and:
          - a
          - c
world:
  dsl: '{stone}'

Trying to load this with e.g. scripts/play.sh -i data/scenarios/bad.yaml results in an error message:

Failed to acquire scenarios from path 'data/scenarios/bad.yaml':
  Parse failure:
    Aeson exception:
    Error in $: Prerequisites graph contains cycles: [a -> b -> c -> d]

It is correct to reject this scenario, but the reported cycle a -> b -> c -> d is bogus: b does not depend on c. In fact, there are two cycles, a -> b and a -> c -> d.

Expected behavior
The error message should report a true cycle, i.e. a sequence of nodes where each depends on the next, and the last depends on the first.

Additional context
stronglyConnComp computes a topologically sorted list of strongly connected components (SCCs). It is true that every component will be a singleton vertex (AcyclicSCC) if and only if the graph has no directed cycles. However, in the case that there are nontrivial SCCs, the list of vertices reported in a (confusingly named) CyclicSCC (or NECyclicSCC) is in no particular order and need not represent a cycle. The above example constructs a graph with four vertices, where a and b are mutually reachable and there is a directed cycle a -> c -> d -> a. In this case all four vertices are part of the same SCC (since any vertex can be reached from any other) but they obviously do not form a cycle. Even in the case where an SCC does actually consist of a cycle and nothing else, I am not even sure it is guaranteed that the vertices will be returned in order around the cycle. It seems a reasonable assumption, but the documentation does not specify.

I plan to fix this shortly but wanted to file an issue for discussion/posterity.