enforce the validity of NIR graphs on creation

[stevenabreu7]

PR #59 introduces code to check the consistency and "defined-ness" of input and output types across a NIR graph (also pointed out in #4). This is currently not enforced.

However, NIR, as an intermediate representation, would do well to enforce the correctness of all created NIR graphs. Otherwise, we end up with potentially many invalid NIR graphs that will effectively be un-parseable.

This PR enforces the validity of NIR graphs, as defined by the NIRGraph._check_types() method. If the graph types are undefined or inconsistent, a warning is printed and an attempt is made to infer them with NIRGraph.infer_types(). If this fails, an error is thrown.

TODO:

• fix tests

Note: this can also be merged only after our paper deadline.

[Jegp]

This sounds like a solid idea, but may I suggest adding a flag somewhere so the user can choose to ignore the pass? Maybe a parameter like ignore_type_check = False?

[stevenabreu7]

Good idea, I've added a flag as NIRGraph parameter (ensure_validity). I've also put a strict flag into the read and write functions for NIRGraphs. This is turned on by default (since people should be using this), but can be turned off.

Many tests are failing now, because we had a lot of inconsistent graph types defined in the tests..