Similar to SERVER-82324, the thread in kill_sessions_common.cpp acquires the client lock and reads the logical session id on opCtx, while the thread in resharding_collection_cloner.cpp sets a new logical session id without the client lock.

See this comment for full context.

It didn't reproduce locally or in a patch, but it seems like we are pretty confident in the problem and solution based on the stack trace and similar issues in the past.