Env vars + partial parsing

[jtcohen6]

If an env var value needed at parse time (input to ref/source/config or yaml property) has changed, and partial parsing is enabled, dbt will not re-parse affected nodes and is likely to yield incorrect results.

Reproduction case: gist

• Non-solution (better than nothing): Continue to document this in known partial parsing limitations; add to env var docs as well. Instruct users to manually trigger a full re-parse (--no-partial-parse) when they change an env var value needed at parse time.
• Simpler solution: Store a dictionary of all available env vars in the internal manifest (partial_parse.msgpack). If any env var values change, trigger a full re-parse, and say that a change in env vars is the reason. Note: I'm not sure how this would work in deployment environments that set contextual env vars, such as dbt Cloud scheduled runs.
• Complex, good solution: For each node in the graph, statically analyze calls to the env_var macro, and then store those env vars in depends_on for that node a separate map object, similar to how we handle docs blocks. Then, if we detect a change in env var values, we can trigger re-parse for only those nodes which depend on that env var, plus (if the node is a macro) any nodes that depend on the macro. (It would be great to do this for --vars someday as well.)

[jtcohen6]

@gshank Up to you whether we keep this issue open, or close it and open a new one for env vars in other files (.yml files, dbt_project.yml, profiles.yml)