[LibOS] Fix shim_ipc_info leak

This PR has one caveat: it makes pids reusable i.e. they might not be
growing (Graphene might reuse old, freed pids).

LibOS/shim/src/bookkeep/shim_process.c

@@ -77,6 +77,9 @@ struct shim_child_process* create_child_process(void) {
 void destroy_child_process(struct shim_child_process* child) {
     assert(LIST_EMPTY(child, list));
 
+    /* This only removes the pid from internal IPC tracking in this process.  */
+    release_ipc_id(child->pid);
+
     free(child);
 }
 

LibOS/shim/src/ipc/shim_ipc_ranges.c

@@ -272,7 +272,7 @@ static int add_ipc_range(IDTYPE base, IDTYPE owner, const char* uri) {
 static void __del_ipc_subrange(struct subrange** ptr) {
     struct subrange* s = *ptr;
     *ptr = NULL;
-    put_ipc_info(s->owner);
+    put_ipc_info_in_list(s->owner);
     free(s);
     nsubed--;
 }

LibOS/shim/src/sys/shim_clone.c

@@ -369,7 +369,9 @@ long shim_do_clone(unsigned long flags, unsigned long user_stack_addr, int* pare
         /* We should not have saved any references to this thread anywhere and `put_thread` below
          * should free it. */
         assert(__atomic_load_n(&thread->ref_count.counter, __ATOMIC_RELAXED) == 1);
-        /* We do not want to release `tid`, now it's owned by child process. */
+        /* We do not want to release `tid` now, we will do this once the child process dies. This
+         * makes little sense (as in theory we've passed the ownership of this id to the child), but
+         * IPC code is a mess, so that's what we have to do. */
         thread->tid = 0;
         put_thread(thread);