Removing the Stage Property of Nodes

[urig]

Hi,

As we (@urig, @qguyk, @galwiner) have learned from examining issue #123, the Entropy API is currently confusing and somewhat even buggy around Nodes, their ids, stages and labels.

Problems

1. The Nodes table in Sqlite has an id column but that is not the primary key for the table.

2. The primary key for the Nodes table is a combination of id and experiment_id. This is a correct natural key for nodes but we prefer synthetic (artificial) keys.

3. Given an experiment, both stage and node id (the column, not the actual natural key combination of experiment_id and the column) are two different names for the same thing.

4. The SqlAlchemyDB method get_nodes_id_by_label() is broken because it returns node ids without experiment_ids. This does not follow the principle of least surprise :)

5. The idea behind stage is that in the context of a specific experiment, users can accidentally (?) use the same label for two different nodes. stage can be used to distinguish between different nodes in this case. I* think this solves a problem we should instead prevent.*

Proposed solution

a. Change the unique identifier for nodes from the id X experiment_id combination to a single auto-generated id column.

b. Remove all references to stage from the Entropy code base and replace them with node_label. That is, the label property of nodes.

c. Make it impossible for users to use the same label for two different nodes in any given experiment. We can enforce this in the
Sqlite DB and reflect it to users when they invoke the Graph constructor.

I think this solution will make our code base and our data schema smaller, more intuitive and easier to reason about.

Would love to hear what you think - Your feedback most welcome!