mro: switch from dfs to c3 as default

[rurban]

The default MRO needs to switch from the old DFS (depth-first-search) to the recommended and perl6/Moose C3 algorithm, which was previously optional. All object systems based on inheritence need to use C3 to avoid the diamond problem. And this change speeds it up, avoiding run-time loading of ext/mro, and avoids having the old dfs in memory.

If it's not possible to fix all the internal and cpan modules, just use C3 for the new classes.

[rurban]

This change needs to export sv_sethek to be visible to CORE extensions. Xp => EXpd

And the ISA order for some classes changed now. Need to adjust some fragile tests, and even some yet unknown c3 bugs (#251)

• run/switchDBG.t for -DC

Inconsistent hierarchy during C3 merge of class 'Math::BigInt::Subclass':
	current merge results [
		Math::BigInt::Subclass,
	]
	merging failed on 'Exporter' at t/Math/BigInt/Subclass.pm line 17

Full list of regressions (~1 or 2 bugs):

	../cpan/Math-BigInt/t/inf_nan.t
	../cpan/Math-BigInt/t/isa.t
	../cpan/Math-BigInt/t/sub_ali.t
	../cpan/Math-BigInt/t/sub_mbf.t
	../cpan/Math-BigInt/t/sub_mbi.t
	../cpan/Math-BigInt/t/sub_mif.t
	../cpan/Math-BigInt/t/upgradef.t
	../cpan/NEXT/t/actual.t
	../cpan/NEXT/t/actuns.t
	../cpan/NEXT/t/unseen.t
	../cpan/libnet/t/ftp.t
	../cpan/libnet/t/require.t

zlib is also broken, but probably related to the attempted broken zlib-1.2.11 update.

[rurban]

Testing CPAN with c3, and submitting tickets. Maybe we need a backcompat Configure setting -Dmro=dfs, but better not.

Fixed problems in CORE:

• NEXT PR#1
• libnet PR#31
• EUMM ~~PR #286~~ (TODO broken on Win32/VMS, use mro cannot be used)
• IO-Compress #120239
• Math-BigInt #120242

On CPAN:

• CPAN-SQLite #120241
• NetAddr-IP
• Text-CSV

Blocked by:

• EUMM http://github.com/Perl-Toolchain-Gang/ExtUtils-MakeMaker/issues/91 (my c3-wip branch)