* use the "yield_waiter" and "waiter" from common/async insteasd of the "waiter"
  implemented inside the bucket notification code (this is so we don't
  need separate investigations for 2 implementations)
* added a unit test that simulate how a separate thread (kafka or amqp) is
resuming a coroutine which is created by either the frontend or the
notification manager.

before using "defer" the unit test is passing, however,
when executed under thread sanitizer (using the WITH_TSAN cmake flag)
the following errors are observed: https://0x0.st/Xp4P.txt
after using "defer" the unit test passes under TSAN without errors.

Fixes: https://tracker.ceph.com/issues/64184

Signed-off-by: Yuval Lifshitz <ylifshit@ibm.com>
(cherry picked from commit 2872c75)

In file included from src/rgw/driver/posix/bucket_cache.h:19,
                 from src/test/rgw/test_posix_bucket_cache.cc:4:
src/common/cohort_lru.h: In member function _void cohort::lru::TreeX<T, TTree, CLT, CEQ, K, LK>::lock()_:
src/common/cohort_lru.h:334:14: error: _for_each_ is not a member of _std_
  334 |         std::for_each(locks.begin(), locks.end(),
      |              ^~~~~~~~
src/common/cohort_lru.h: In member function _void cohort::lru::TreeX<T, TTree, CLT, CEQ, K, LK>::unlock()_:
/home/yuvalif/ceph5/src/common/cohort_lru.h:339:14: error: _for_each_ is not a member of _std_
  339 |         std::for_each(locks.begin(), locks.end(),
      |              ^~~~~~~~

Signed-off-by: Yuval Lifshitz <ylifshit@ibm.com>