Investigate performance of `matches_any_publishers()` when enabling intraprocess

[mini-1235]

Related to https://discourse.openrobotics.org/t/performance-characteristics-subscription-callback-signatures-rmw-implementation-intra-process-communication-ipc/51454/9?u=mini-1235

From the flamegraphs, it seems like rclcpp::experimental::IntraProcessManager::matches_any_publishers has a non-trivial overhead, it would be interesting to investigate whether this can be optimized

Opening this issue to make sure it's not forgotten, more background and discussion can be found in the original post linked above

[fujitatomoya]

@mini-1235 #3068 works for you?

[mini-1235]

@fujitatomoya Thanks for the super fast PR! I will test it and get back to you in a few days

[fujitatomoya]

I tried to address this performance improvement with #3068 and #3073, but it appears that it generates the regression for Nav2 stack with segmentation fault. core stack trace obviously tells me something wrong in rclcpp hash map that i introduced by those PRs.

after all, #3077 will revert #3068, until we figure out what the root cause is and keep the stability for userland.

[fujitatomoya]

@mini-1235 just a question, did you rebuild everything including nav2 pacakges with #3068? (reference: #3068 (comment))

[fujitatomoya]

i tries to fix this issue with #3084 again, all tests are passing for rclcpp.

[mini-1235]

@mini-1235 just a question, did you rebuild everything including nav2 pacakges with #3068? (reference: #3068 (comment))

Yes, I also added some additional logs during testing, and I can confirm that they are being printed :)