Merge bitcoin/bitcoin#27116: doc: clarify that LOCK() internally checks whether the mutex is held

91d0888921 sync: unpublish LocksHeld() which is used only in sync.cpp (Vasil Dimov)
3df37e0c78 doc: clarify that LOCK() does AssertLockNotHeld() internally (Vasil Dimov)

Pull request description:

  Constructs like

  ```cpp
  AssertLockNotHeld(m);
  LOCK(m);
  ```

  are equivalent to (almost, modulo some logging differences, see below)

  ```cpp
  LOCK(m);
  ```

  for non-recursive mutexes, so it is ok to omit `AssertLockNotHeld()` in such cases. Requests to do the former keep coming during review process. `developer-notes.md` explicitly states "Combine annotations in function declarations with run-time asserts in function definitions", but that seems to be too strong or unclear. `LOCK()` is also a run-time assert in this case.

  Also remove `LocksHeld()` from the public interface in `sync.h` since it is only used in `sync.cpp`.

ACKs for top commit:
  achow101:
    ACK 91d0888921
  hebasto:
    ACK 91d0888921, I have reviewed the code and it looks OK.

Tree-SHA512: c4b7ef2c0bfeb28d1c4f55f497810f629873137e02f5a92137c02cb1ff603ac76473dcd2171e594491494a5cb87b8c0c803e06b86f190d4acb231791e28e802d

doc/developer-notes.md

@@ -950,7 +950,9 @@ Threads and synchronization
     internal to a class (private or protected) rather than public.
 
   - Combine annotations in function declarations with run-time asserts in
-    function definitions:
+    function definitions (`AssertLockNotHeld()` can be omitted if `LOCK()` is
+    called unconditionally after it because `LOCK()` does the same check as
+    `AssertLockNotHeld()` internally, for non-recursive mutexes):
 
 ```C++
 // txmempool.h