Give date/time/datetime distributions some love

[vankesteren]

• Add non-uniform distributions, e.g., (truncated) normal
• Add von Mises distribution for time
• Add constant interval distributions (day, week, month, year)? Example: column 1 of RKI COVID data
• Look at jsonschema support for datetime values rather than string
• Make sure we test properly

By the way, I saw that in jsonschema we can also put comments like this; might be nice to have when we work on the schema a bit more.

"lower": {
  "type": "string",
  "format": "date",
  "description": "We expect yyyy-MM-dd"
}

[vankesteren]

It may be nice to implement a

class BaseDTDistribution(BaseDistribution):

which deals with translation from underlying distribution to date/time/datetime domain using a particular precision. For example, in generation, this needs to sample a float and multiply by the timedelta of the correct precision (e.g., days).

We may want a

class BaseDTFitter(BaseFitter):

as well then.

Maybe these can even be mixins