hp malloc: differentiate the pkg and shm statistics code

When HP_MALLOC is defined:
	* shm statistics are based on 32/64-bit atomic counters
		- the former logic is still available with HP_MALLOC_FAST_STATS
		- however, stat aggregation and the shm threshold checks should be done by a single process
	* pkg statistics are implemented similarly to the other allocators
	* also re-introduce fragment coalescing for pkg memory

(cherry-picked from commit 268955b)

main.c

@@ -1481,6 +1481,8 @@ int main(int argc, char** argv)
 		goto error;
 	}
 
+	init_shm_statistics();
+
 	/*init timer, before parsing the cfg!*/
 	if (init_timer()<0){
 		LM_CRIT("could not initialize timer, exiting...\n");