create suite-wide metadata cache, caching dataset-level statistics/metadata

[jqnatividad]

Currently, stats compile field/column statistics and persists these stats to a cache file.

This cache file is used by the stats command to return stats instantaneously if the CSV has not changed.
Other "smart" commands also use the stats cache to work faster & smarter.

qsv should have a suite-wide metadata cache that compiles dataset-level statistics and metadata like:

• record-level stats/metadata
  • record width (max, min, avg, median, variance, stddev, mad) and remove the count --width option
• package-level stats/metadata
  • number of duplicate records, which is compiled by the existing sortcheck command, and added to a CSV's stats cache when sortcheck is executed. If a CSV has not changed and sortcheck is executed again, it will fetch the existing duplicate record count in the cache
  • data dictionary as initially inferred by describegpt. Will have a flag to indicate if the data dictionary has been manually curated to prevent auto-updates by future runs of describegpt. If the dataset changes, this flag is reset.