Summarize date columns

[ccunningham101]

In the dataset-report notebook, date columns are summarized with describe() and a bar plot

One use case is for identifying impossible dates. 1900-01-01 is a common impossible date, but impossible dates may differ between data sources. Or checking that dynamic dates, particularly when extracted monthly, fell within an expected range.

The bmi report groups dates by year and then redacts and rounds. This is helpful for identifying common systematic errors (such as a null date as 1900-01-01), rather than individual outliers or issues that may require use of another data column (such as diagnosis before someone's birth date).

Initial implementation in dataset-report format here [Note: the redaction should probably include 0's]

To consider:

1. The ideal grouping time granularity (Y/M/D) probably is some balance of what gives the most information without everything being redacted; could be specified by the user, but may differ between variables
2. Are there potential disclosure issues if multiple variables are querying the same codelist in similar but slightly different ways? Should this be covered by rounding?