Consider atomic counters in internal/otelarrow/admission

[jmacd]

As @lquerel suggested here, open-telemetry/opentelemetry-collector-contrib#36140 (comment), we may wish to reduce the potential for lock contention.

The initial contents of the PR contain an admission2 package which is derived from a well-tested internal code base (with permission). Because the code is well tested w/ Mutex locking, I want to pursue merging it and replacing the original admission (which is FIFO and has a race condition) and the new admission2 (which is LIFO and addresses the race condition).

After the race condition and FIFO->LIFO change is applied, it will be easier to study and test the contention topic exposed here.