[TASK]: Provide better feedback when a model may be problematic

[mwdchang]

Essentially, models extracted from literature are prone to the problem of mistakes in the literature. Users spent an inordinate amount of time debugging the models when "we" could do some of these checks ahead of time.

We have some notion of a model semantic checker in place (in our engineering UI), but it is not leveraged in the canonical application itself because we didn't want to shower the users with warnings/errors. This sentiment has changed in light of the observations above. So we should find ways to detect and surface critical errors to the modellers, so at least they are aware of what directions they need to go.

As a brief, our current checker does a combination of structural and light semantic checks, eg:

• rate laws and units are defined
• declared distribution and actual values are not contradictory
• duplication checks

New want snippets from @liunelson

• Rate law simplification is a MIRA function; in the eval, the users created a lot of summed rate laws (e.g. many conversions in a single transition); it was easy to do it like that for them; but the final model does need to be "simplified" before doing stratification etc.
• tangling production/degradation transitions, that's just counting; they can happen when the terms of a conversion don't match in the equations due to typo
• non-conservative system, allowed but the users should be aware. The eval team encountered this problem.