sample: add time-series sampling options

[jqnatividad]

A lot of data hosted in data catalogs are time-series data.

Since we're focusing on just compiling high-quality, high-resolution metadata in the data catalog, we compile summary stats and frequency tables using the "complete" dataset, but only want to host a representative sample in the catalog while pointing to the source where the "complete" dataset is available.

We don't want the catalog to double as a central datastore with its attendant high capacity reqts, so we only store a sample preview.

However, if we just get the first N rows of a time series dataset, it will most likely always be the same as time-series datasets are often sorted.

Add several time-series sampling options to make the sample more dynamic:

• time-based systematic sampling - ability to specify sampling windows/intervals (1 hour, daily, weekly, etc).
• enhanced starting point - ability to start from the end (most recent observations)
• adaptive sampling/seasonal awareness - e.g. sample more frequently during business days/hours, weekends, seasons, months
• aggregation options - aggregate data within each interval to downsample high-frequency observations