add support for velocity mosaics from granule stacks without a common reference date

[asjohnston-asf]

No description provided.

[forrestfwilliams]

When this function is called we're not currently guaranteed to have all granules between two dates. For example, if we're trying to get velocity between date1 and date5, there might be granules like:

date1_date2, date1_date3 date3_date4, date4_date5

but calling find_needed_granules here with the minmax strategy would only load date1_date3, and date4_date5.

Two possible ways to accomplish this:

1. All ways download all granules between two dates, even if some granules will not be needed
2. Dynamically find the set of granules needed to reconstruct the full deformation history and load them so they can be used in align_to_common_reference_date

[asjohnston-asf]

The two core use cases I see brewing have been:

For a particular frame...

• ...give me a single data band representing net/cumulative displacement from Date A to Date B. This satisfies the needs of the cumulative displacement and secant velocity visualizations.
• ...give me a time series of data bands, each representing net/cumulative displacement from a common reference date, for all available time steps from Date A to Date B. This satisfies the needs of the velocity-via-linear-regression visualization.

I suspect these are also the data-preparation steps data users will need to do for most analysis using this data set. I'm somewhat eager to revisit find_needed_granules, update_reference_date, and align_to_common_reference_date to provide a better interface for those two more generic use cases.

[asjohnston-asf]

I suppose the latter time-series approach satisfies all use cases, but when you only care about total change the former approach would be faster and require less data access.