IO read may use wrong cardinality for property

g.io(...).read() might lose list/set properties. See the example below:

gremlin> graph = TinkerGraph.open()
==>tinkergraph[vertices:0 edges:0]
gremlin> g = graph.traversal()
==>graphtraversalsource[tinkergraph[vertices:0 edges:0], standard]
gremlin> v1 = g.addV().property("feature0", "0.0").property("feature0", "1.1").next()
==>v[0]
gremlin> g.V().valueMap()
==>[feature0:[0.0,1.1]]
gremlin> g.io("graph.json").write().iterate()
gremlin> g.V().drop()
gremlin> g.io("graph.json").read().iterate()
gremlin> g.V().valueMap()
==>[feature0:[1.1]]

By verifying "graph.json", I am sure the write() step works fine. The problem is with read() step.

In https://github.com/apache/tinkerpop/blob/5fdc7d3b5174f73475ca1a48920d5dec614ffc0e/gremlin-core/src/main/java/org/apache/tinkerpop/gremlin/structure/util/Attachable.java#L294-L298, it relies on the host graph to decide the cardinality for the given property key. In TinkerGraph, `features().vertex().getCardinality(anything)` always returns default cardinality SINGLE (unless otherwise configured), which means all vertex properties are created with SINGLE cardinality, even if the graph file itself contains multiple values for that property, as shown in the above case.

I presume TinkerGraph is not the only one who suffers from this problem. For example, for JanusGraph, if the default automatic schema maker is enabled, and a property wasn't defined explicitly, then SINGLE cardinality is returned by default. In this case, the loaded graph will be "wrong" in the sense that it turns LIST/SET values into a SINGLE value.