Add finite constraint for numbers or `number.nanValues`, `number.infValues`

[micimize]

Similar to #393, but proposing a constraint rather than a blanket policy.
I think it should be { "finite": true }. There are many problem domains where NaN, INF, -INF are undesirable.

Also related: frictionlessdata/tableschema-py#249

[roll]

I would even make it stricter having InF/Nan disabled by default with an option (instead of the constraint) like:

number.<allow Inf/Nan - not sure regarding the prop's name> = True
...
number.bareNumber = True
...

[nichtich]

The lexical values of NaN, +Inf, -Inf may also differ. We have trueValues and falseValues for boolean, so nanValues, positiveInfValues, negativeInfValues could be a good idea.

Note that XML Schema datatypes float and double include NaN, +Inf, -Inf but other numeric datatypes don't.

[micimize]

@nichtich that's an excellent idea as well, though I'd suggest having both as options as having to specify three empty lists for the common finite case feels a bit fragile, even if it means a mutual exclusion validation.

[roll]

I think nanValues and infValues are really nice proposal! It's not supported by pandas or polars but can be easily added on the post-processing stage.

Currently, I'm removing it from v2. Let's first get a demand metrics using this instrument - #863