Fix: Robustness for dates and missing values in aligned dataframes which are commonly produced from OMOP

[eroell]

Fix #303

[Zethson]

Should this be solved on this level or on the OMOP reader level?

[eroell]

Tend to fix here.
There's also other ways to write dates into ehrdata's dataframes. And they are also useful to treat as dates, for e.g. sorting.
Writing these objects is somewhat annoying. I have a bunch of "for column in edata.obs" lines in my analyses already, but did not yet want to commit having this a fixed way of writing. That OMOP will usually fail to write convinces me that this should work without user workarounds now