Customers are often baffled by dropped points which fall outside a retention policy. Often the cause is writing with the wrong precision, but short-lived shards can also cause this error to occur.

This proposal is to log the earliest rejected point in a write request and the latest rejected point in a write request.

In 1.X the new error handling should also distinguish rejection for write window and rejection by retention policy.

For printing the points, we should include the bounding time that caused the rejection, the point's timestamp, and the database and retention policy, and the point's key (measurement and tag values). That's constrained in length (no 2K string fields), keeps field values out of the log, but provides identifying information for the point.

This will require changes to the tsdb.PartialWriteError for computing and printing the dropped points summary.