Increase in execution time of `mojo::IncomingInvitation::Accept`

[Andrew Moylan]

Hi, in the ml_service on ChromeOS, starting in M144, we are seeing a sudden ~2x to ~10x increase in execution time of `mojo::IncomingInvitation::Accept`.

This was noticed by a Finch alert on our metric MachineLearningService.WorkerProcessAcceptMojoConnectionTime here.

(Internal bug link with graph & links: b/483019693)

That code runs in some worker processes that ml_service daemon spawns. The worker processes accept invitations from the main ml_service daemon process there

The increased execution time should not cause any issue on ChromeOS ml_service, but I just wanted to raise it in case it is unexpected / could be indicative of some other issue. Did something change in M144 that could reasonably cause this?