Say that points arrive per source

[kkdev92]

A person with two devices gets the same measurement from both, so ListAsync returns it twice
and summing the points inflates the total. Nothing in the response indicates it: every point is
genuine, and the count is only wrong once the set is treated as one person's day.

Filtering by Platform does not separate them either — a phone's built-in tracking and a wrist
device from the same vendor report the same one. Device.DisplayName is the field that does.

docs/data-points.md gains a section that says this, points at the roll-up operations for a
figure, and records what Reconcile gives up in exchange for merging: a ReconciledDataPoint
carries no dataSource, so it answers what the value is and not which device produced it. It
also names the point DataPointName rather than Name.

Documentation only — no change under src/.

Not included

Three things were considered and left out:

• The roll-up range caps. Already on the Range property of both roll-up requests, which is
  where you are looking when you set one. Repeating them here would be a second copy to keep.
• The roll-up response array name. RollUpDataPointsResponse exposes RollupDataPoints and
  nothing else, so a typed caller cannot reach for the wrong one.
• Per-type page density. The general rule under Following the cursor can return fewer records
  than exist already covers it, and covers it better than a list of types would.

Every claim in the new section was checked against the generated models rather than written from
recollection.

🤖 Generated with Claude Code