THX_ck_entersub_args_sereal_decoder(): set parent link correctly

This patch was originally provided by Dave Mitchell inline as a response
to a "Blead Breaks CPAN" report in the Perl 5 bug queue (originally
rt.perl.org; now github.com/Perl/perl5/issues).  Dave's comment in
original ticket.

    THX_ck_entersub_args_sereal_decoder() in Decoder.xs
    wasn't setting the parent link correctly when munging an op
    tree.  The following diff makes all tests pass on blead. I
    haven't tested against older perls.

References:

Perl/perl5#17114
(originally https://rt-archive.perl.org/perl5/Ticket/Display.html?id=134320)

Sereal#207

Perl/Decoder/Decoder.xs

@@ -223,13 +223,17 @@ THX_ck_entersub_args_sereal_decoder(pTHX_ OP *entersubop, GV *namegv, SV *ckobj)
     if (arity > min_arity)
         opopt |= OPOPT_OUTARG_HEADER;
 
-    OpMORESIB_set(pushop, cvop);
-    OpLASTSIB_set(lastargop, op_parent(lastargop));
+    /* cut out all ops between the pushmark and the RV2CV */
+    op_sibling_splice(NULL, pushop, arity, NULL);
+    /* then throw everything else out */
     op_free(entersubop);
-    newop = newUNOP(OP_NULL, 0, firstargop);
+    newop = newUNOP(OP_NULL, 0, NULL);
     newop->op_type    = OP_CUSTOM;
     newop->op_private = opopt;
     newop->op_ppaddr = opopt & OPOPT_LOOKS_LIKE ? THX_pp_looks_like_sereal : THX_pp_sereal_decode;
+    /* attach the spliced-out args as children of the custom op, while
+     * deleting the stub op created by newUNOP() */
+    op_sibling_splice(newop, NULL, 1, firstargop);
     return newop;
 }