get_object_locking carries the freshness stamp its latest-anchored read owes, and the shrink-only census shrinks back (Erik's ruling on #3878's roster call) #3880

Erik's ruling on the judgment call recorded in #3878 / PR #3879: stamp it, don't roster it.

#3879's fix made DarlingObjectStatsReader.IndexLockingSql visible to the latest-anchored-reads census for the first time (its old per-name anchor sat inside a CTE, which the scanner ignores as picking a set — the defect itself had hidden the read). The lane rostered it on the shrink-only LatestAnchoredReadsWithoutTheirStamp list with the pre-existing-debt-made-visible argument. Erik prefers the better end state: the read gains a real freshness stamp and the roster shrinks back.

Work:

• get_object_locking's payload gains the house freshness stamp for latest-anchored reads — mirror exactly how the already-stamped snapshot reads in DarlingObjectStatsReader.cs / the McpLatestSnapshotStampTests swept set carry theirs (field name, semantics, and placement come from the existing pattern, not invented; the collection_health_age_seconds family is the reference for age-style stamps if that is the house shape here).
• Remove the IndexLockingSql entry Darling's four object-locking reads anchor on the server's latest capture, not on per-name groups (#3878) #3879 added to LatestAnchoredReadsWithoutTheirStamp in McpPayloadContractCensusTests — the census resumes shrinking, which is its designed direction.
• McpLatestSnapshotStampTests' sweep already includes GetIndexLockingAsync; verify the read now satisfies the stamped arm rather than the roster exemption.
• Tool Description (and the get_object_locking instructions row, if one exists) documents the field the way the stamped siblings do.
• Cross-SKU parity check: if Lite's get_object_locking twin (Lite/Mcp/McpObjectStatsTools.cs) is the same latest-anchored-unstamped shape, stamp it identically; if Lite's censuses have no such roster, note the asymmetry rather than inventing one.
• The Darling's four object-locking reads anchor on the server's latest capture, not on per-name groups (#3878) #3879 live rename pin (DarlingIndexLockingRenamedDatabaseLivePostgresTests) parses the tool envelope — extend it (or the surface pin) to assert the stamp is present and sane.

Follow-up to #3876 → #3877 (Lite fix), #3878 → #3879 (Darling twin + the roster call this reverses by ruling).