fix(v4): a killed node's children become Hera's; Review Focus 4 checked

From the final review of MESH step 7.  NODE-KILL left children whose parent
was the dead number, so a node born later under it inherited authority over
them; they are re-parented to Hera.  After UNIT-KILL Hera no longer lists the
removed centre, and a later KILL completes (shown to fail with GONE-DOWN's
unlisting taken out).  test_manage 45/45 both widths; test_host_unit 127/127.

Acceptance: amd64, aarch64, riscv64 clean qemu: ok>, POST 538/538,
dict_hash=0x248b4edc47f4357e on all three; test-32 0 failures.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

v4/tests/test_host_unit.c

@@ -717,6 +717,11 @@ int main(void)
     CHECK(node_numbered(10)->mem[LOST] == hera_lost && node_numbered(11)->mem[LOST] == 0 && node_numbered(13)->mem[LOST] == 0 && node_numbered(14)->mem[LOST] == 0
           && waiting(10) && waiting(11) && waiting(13) && waiting(14), "nothing was let go in all of this, and all wait again: Hera %ld (%ld before)",
           (long)node_numbered(10)->mem[LOST], (long)hera_lost);
+    /* Review Focus 4 (the final review of step 7): after UNIT-KILL no node still lists a removed node as its
+     * child -- the next KILL would tell it, and with no way to it that would be an error. */
+    CHECK(strcmp(tell(10, "20 (WHICH) ."), "-1 ") == 0, "Hera no longer lists node 20 among the nodes she birthed: \"%s\"", printed);
+    CHECK(told(10, "14 KILL") && node_numbered(14) == NULL && strcmp(tell(13, "1 2 + ."), "3 ") == 0,
+          "and a KILL after UNIT-KILL tells only the nodes there are: it completes, and the rest go on: \"%s\"", printed);
 
     printf("  %d checks, %d failures\n", checks, failures);
     return failures != 0;