API: Rm default value from time_delta for cd_index.

[rossbar]

I'd like to propose a change to the cd_index API: removing the default value for time_delta. The current default of 5 years as a timedelta object feels arbitrary, and I can't think of a strong reason why a default value makes sense here. Furthermore, providing a datetime.timedelta object suggests that the time node attribute must be datetime.datetime objects by default, which feels like too strong a recommendation IMO.

[MridulS]

I agree we should let users decide this bit. We should also update the error raised in line 106-108, there is no default datetime type anymore.

[rossbar]

We should also update the error raised in line 106-108, there is no default datetime type anymore.

Great catch! Should be fixed in 4048403.

[dschult]

The failures here are not related... And geospatial example somehow doubled the size of its array.

And because I tried to use a github comment "suggestion" instead of just pushing to this PR, ended up with two commits instead of one and my test was hidden (unless you click "show resolved"). So here's that text:

I'm wondering if we could add an example that doesn't use datetime. I tried quickly to take this example and make the dates just a year as an int e.g. '2012'. To get it to work, I had to make time_delta=4 instead of time_delta=5. Did I miss something? Or maybe there is an "off-by-one" error in the 5*365 value (or in how time_delta is used).

[dschult]

I think this is ready for merging (or at least review). It improves the current interface and improves the docs.