Consumable retention policy

[Stiivi]

Define consumable retention policy. Currently the retention is expected to be provided by the object, which is in most of the cases sub-optimal such as consuming all data into list of Python objects. Also this implementation is not aware of context in which the node is being executed. Suggestion:

1. ExecutionEngine subclasses might insert retention/caching nodes after consumables. Advantage: simple implementation, aware of broader processing context. Disadvantage: might not be backend aware.
2. Retention operations: retain(object, times) Advantage: backend aware. Disadvantage: not aware of the broader processing context.