new sub and signature parser leak on errors

[rurban]

PERL_DESTRUCT_LEVEL=2 valgrind --leak-check=full ./miniperl -Ilib -e'+sub:a{}'

With GH #7 , the interim signature parse leaks the lex op yyval.opval on errors.

==10628== 222,208 (7,168 direct, 215,040 indirect) bytes in 14 blocks are definitely lost in loss re
cord 43 of 43
==10628==    at 0x4C2AD10: calloc (vg_replace_malloc.c:623)
==10628==    by 0x415CE1: S_new_slab (op.c:181)
==10628==    by 0x415DED: Perl_Slab_Alloc (op.c:226)
==10628==    by 0x4239A1: Perl_newOP (op.c:4748)
==10628==    by 0x4A0FED: S_pending_ident (toke.c:8257)
==10628==    by 0x486A02: Perl_yylex (toke.c:4340)
==10628==    by 0x4B4F04: Perl_yyparse (perly.c:322)
==10628==    by 0x63CAD9: S_doeval (pp_ctl.c:3484)
==10628==    by 0x642B49: Perl_pp_require (pp_ctl.c:4169)
==10628==    by 0x52BB2B: Perl_runops_debug (dump.c:2232)
==10628==    by 0x40B4D4: Perl_call_sv (perl.c:2740)
==10628==    by 0x414117: Perl_call_list (perl.c:4846)

testcase: cperl t/op/svleak.t

not ok 130 - OP_SIGNATURE
# Failed test 130 - OP_SIGNATURE at t/op/svleak.t line 32
#      got "1"
# expected <= "0"
not ok 131 - OP_SIGNATURE err
# Failed test 131 - OP_SIGNATURE err at t/op/svleak.t line 32
#      got "1"
# expected <= "0"

This only fails in the errorcase.

[rurban]

Only leaks in the error case.

[rurban]

Ok, this is a wrong test, which assumes it will leak one more sv as multideref, but does not.
Fixed the test with 99190cbfbd7006166c68a6268b78f9ed65e3e3e9