[mini-1235]

Description

Fixes #1086.

Buffer-aware publishers and subscriptions add themselves to the graph cache with parse_put after creation, and their liveliness token adds them again, as tokens from the same session also reach the graph cache. The entity sets deduplicate by GID, so graph queries were correct, but matched events (and the buffer discovery callbacks) were reported on both puts, while a removal is reported once. update_topic_map_for_put now ignores a put for an entity that is already in the graph.

Test: ros2/rmw_implementation#307.

Is this user-facing behavior change?

Yes. Matched events for types with uint8[] fields report the actual counts.

Did you use Generative AI?

Yes, Claude Code (Claude Opus 5.5) wrote the fix and this description. Reviewed and tested by me.

Additional Information

One publisher and two subscriptions, then one subscription removed, then a second publisher added and removed; a plain type (int8[]) and a uint8[] type with cpu and any subscriptions, either creation order, in the same process and across processes:

	rolling	this PR
publisher current/total with 2 subscriptions, same process	4/4	2/2
publisher with 2 subscriptions, across processes	3/3	2/2
first subscription after a second publisher	4/4	2/2
uint8[] type compared with the plain type (cpu/any × creation order × same/other process)	differs in all 8	same in all 8

test_rmw_implementation passes. With a uint8[] field added to test_msgs, test_event (4 failures) and rclcpp's TestQosEvent (4 matched-event failures) pass as well.

🤖 Generated with Claude Code