add-explore-record-buckets

Explore's over-time chart currently fans out from the console to many per-stream aggregateRecordsByTime calls. That makes first load slow and keeps the chart sparse when the selected corpus spans years because granularity is derived from a fixed client window instead of the populated record extent.

The existing Explore timeline endpoint already defines the authoritative owner-visible record set across (connectorinstanceid, stream) partitions. The bucket endpoint uses the same set semantics and the same semantic timestamp expression, but returns aggregate counts instead of record bodies.

The reference implementation SHALL expose GET /ref/explore/records/buckets as an owner-session-authenticated reference route that returns exact bucket counts across the same owner-visible record set used by GET /ref/explore/records. The route SHALL NOT be reachable over /mcp or with a grant-scoped token.