[Update] Best practices for backfilling large data with caggs

[mfreed]

If you are creating a cagg on a hypertable with a lot of existing data, current best practices are not to create a policy back to beginning of time, but instead create a standard refresh policy (e.g., for last week or so), then use (potentially several calls to) refresh_continuous_aggregate to catch it up the initial time.

We should add this best practices to docs around continuous aggs