`aggregate.time_aggregate` incorrectly handles aggregating over months and years

`aggregate.time_aggregate` converts its aggrigation interval from an ISO duration string into a total number of seconds. This runs into issues if it contains months or years, as they are not a consistent number of seconds.

For example, a period of 1 month starting on 2000-01-01 would give:
2000-01-01 -> 31 day difference -> 2000-02-01 -> 28 day difference -> 2000-03-01

Currently this different length is not taken into account.

Start by locating aggregate.time_aggregate and reviewing how its ISO duration string is converted into seconds. Reproduce the month-based example from the issue, then ensure aggregation accounts for the varying calendar lengths of months and years and add coverage for those intervals.

Written by the indexing model from the issue text.