Expose info about attribute coercion and/or way to validate user data/layout

[etpinard]

Two ways of doing this come to mind:

1. Add a few Lib.log in coerce.js notifying users if a given attribute is set to an invalid value. Post Log levels #590, these logs would be hidden by default. This method has one drawback: it would only warn user about invalid values, not about invalid attributes (which never reach coerce.js.
2. Add a Plotly.validate(data, layout) method that runs the full supplyDefaults chain, compare input and full data/layout and logs the info.

[cpsievert]

As mentioned in #619, having a required meta tag for attributes would be quite useful for generating informative error messages in our client APIs

[etpinard]

I just pushed some Proof-of-Concept commits for a Plotly.validate method to plotly-validate.

I wouldn't mind getting opinions from @cpsievert @mdtusz @alexcjohnson @chriddyp @cldougl

[etpinard]

Note that the idea of providing an error code comes from the deep-diff package.

I'm imagining that for some applications e.g. the unsettable validate error code would not matter.

[alexcjohnson]

@etpinard I love the strategy and the structure of the response. Aside from a few comments I made on the commits, I would just try and make the error codes look more like errors than benign "differences" like those returned by deep-diff. So instead of "visible", use "invisible" or even "aborted" because that's likely what happened - coercion was aborted because something crucial was missing. "unsettable" I might just call "unused", and it needs more explanation like "key XXX is valid but was not used because other values made it irrelevant"

[mdtusz]

Looks pretty good. Github won't let me comment on the code for some reason, but it may be nice to pass a list of the path keys rather than just root to crawl. E.g. ['xaxis', 'rangeslider', 'range'] so that if there arises any specific validation based on the nest, we can test it.

[alexcjohnson]

You can't comment on the combined diff until it's in a PR I guess... But you can comment on individual commits