Expose pushProfileTimer / pop ProfileTimer in PhysicsClient API to benchmark Python parts of PyBullet.

reduce 'm_cooldownTime' from 1000 microseconds to 100 microseconds (overhead in raycast is too large)
If needed, we can expose this cooldown time.
Replace malloc by btAlignedObjectArray (going through Bullet's memory allocator)

src/BulletDynamics/ConstraintSolver/btSequentialImpulseConstraintSolver.cpp

@@ -781,7 +781,8 @@ int	btSequentialImpulseConstraintSolver::getOrInitSolverBody(btCollisionObject&
     else
     {
         // Incorrectly set collision object flags can degrade performance in various ways.
-        btAssert( body.isStaticOrKinematicObject() );
+        //btAssert( body.isStaticOrKinematicObject() );
+        //it could be a multibody link collider
         // all fixed bodies (inf mass) get mapped to a single solver id
         if ( m_fixedBodyId < 0 )
         {

src/LinearMath/TaskScheduler/btTaskScheduler.cpp

@@ -412,11 +412,13 @@ static void WorkerThreadFunc( void* userPtr )
             }
         }
     }
-
-    // go sleep
-    localStorage->m_mutex.lock();
-    localStorage->m_status = WorkerThreadStatus::kSleeping;
-    localStorage->m_mutex.unlock();
+	{
+		BT_PROFILE("sleep");
+		// go sleep
+		localStorage->m_mutex.lock();
+		localStorage->m_status = WorkerThreadStatus::kSleeping;
+		localStorage->m_mutex.unlock();
+	}
 }
 
 
@@ -503,7 +505,7 @@ public:
             storage.m_threadId = i;
             storage.m_directive = m_workerDirective;
             storage.m_status = WorkerThreadStatus::kSleeping;
-            storage.m_cooldownTime = 1000; // 1000 microseconds, threads go to sleep after this long if they have nothing to do
+            storage.m_cooldownTime = 100; // 100 microseconds, threads go to sleep after this long if they have nothing to do
             storage.m_clock = &m_clock;
             storage.m_queue = m_perThreadJobQueues[i];
         }