Functional:
  David H. discussion about JDK-8230249:
  - Use ref_counts to protect ObjectMonitor* cached in JavaThread
    current_pending_monitor and current_waiting_monitor fields
  - Update current_pending_monitor() and current_waiting_monitor()
    callers to use ObjectMonitorHandles and new set_om_ptr_if_safe()
  David H. 1 of 3 CR7 review:
  - Remove unnecessary load_acquire() and release_store() calls from
    the counter fields.
  Robbin CR7 review:
  - Add mark_next_for_traversal() for implementing safe traversal of
    lists by concurrent readers.
  - Update ObjectMonitor audit and checking functions to do safe
    traversal of the lists where needed.
  David H. 3 of 3 CR7 review:
  - owner_is_DEFLATER_MARKER() and ref_count no longer use load_acquire().
  - Rename three param set_owner_from() to simply_set_owner_from() and
    drop cmpxchg() so the set of the _owner field to new_value happens
    without any memory sync.
  - Add two param simply_set_owner_from() that sets the _owner field to
    new_value without any memory sync.
  - Rename set_owner_from_BasicLock() to simply_set_owner_from_BasicLock(),
    drop "assert(is_lock_owned(basic_lock_p))" since all the callers
    already check that.
  Erik O. bug fix:
  - The BasicLock needs to be initialized to markWord::unused_mark before
    C2 inc_om_ref_count() can take a branch to DONE_LABEL for the slow path.
  David H. CR7 follow-up:
  - mark_next() does not need load_acquire(&_next_om) since it is followed
    by cmpxchg(); regular load will do.
  - g_block_list is only changed by cmpxchg() so load_acquire(&g_block_list)
    is not needed; missed these from a previous David H. comment.
  Self-review:
  - Verify that all primary list head changes are made with mark_list_head()
    which uses cmpxchg(); list head loads can be done with regular loads
    instead of load_acquire() and all unmarking of list heads can be done
    with a regular store followed by an OrderAccess::storestore() instead
    of a release_store().

Test update:
  - no changes

Collateral:
  - Drop 'ObjectMonitorHandle(ObjectMonitor*)' ctr and use new
    set_om_ptr_if_safe() in monitors_iterate()
  - Drop 'ObjectMonitor::is_active()' and use '!ObjectMonitor::is_free()'
    directly.

Cleanup:
  - Add spaces around a few binary operators.
  - ObjectSynchronizer::finish_deflate_idle_monitors() should do nothing
    with async deflation unless a special deflation has been requested.
    - restore finish_deflate_idle_monitors() to mostly match the baseline
    - audit_and_print_stats() is now called from do_safepoint_work()
      after async deflation is requested.
    - deflate_idle_monitors_using_JT() now logs the global list counters
      at the Info level for async deflation.
  David H. 3 of 3 CR7 review:
  - Update assert()'s and guarantee()'s that refer to ref_count() to save
    a local copy of the value, do the check with the local copy and report
    any failure with the local copy and the current ref_count() value for
    comparison.
  - Update (monitorinflation, owner) log messages to provide more caller
    context.
  - Restore "assert(this->object() != NULL)" in ObjectMonitor::enter()
    to the baseline version.
  - Add a clarifying comment in is_busy_to_string().
  - Delete stale baseline comment in ObjectMonitor::exit().

Temporary:
  - no changes