Don't use ToEventStream as inputs

[CJ-Wright]

Currently when we walk up the graph we then walk down to check if the node has a ToEventStream subscribed to the node. If so we use that ToEventStream as our input uids allowing us to only cache that part of the pipeline. However this assumed that data which came out of a ToEventStream node was stored somewhere, which might not always be true. We could inspect if the ToEventStream node has a db.insert decedent, but that might be a bit brittle. Additionally this very funny walk might be more computationally expensive than calling nx.ancestors and looking for FromEventStream instances. (Note that this might be able to be mitigated by playing some fun graph games, maybe by searching the graph for all FromEventStream nodes and all ToEventStream nodes, looking at the paths to those nodes and finally making decisions based off the path (eg is the ToEventStream attached to a node on the path?)

tl;dr our magic for ToEventStream input tracking is nice, but maybe too smart for its own good. We might remove it to improve init time, simplicity, and developer sanity.