Constexpr conversion to MPL types

[Kojoley]

Fixes #133

GCC, Clang, MSVC, ICC, all of them are fine with this https://godbolt.org/z/6K6NDm

[jzmaddock]

As noted in #133 I'm rather inclined to treat this as a non-bug, or at least one that can't be fixed without breaking stuff.

The problem with this fix is it means that code that relies on this conversion, and on mpl::bool_ being at least forward declared after including a type_traits header will be broken (and will need to be modified to include the mpl headers explicitly, as you did for the test case, which quite deliberately did not do that).

I realise they should be doing that already, but if we're forcing folks to change their code, it would be better to mark this interface as deprecated, and suggest they dispatch on integral_constant instead.

[Kojoley]

Ok, I just will create a new duplicate test that and will not touch this to show that nothing should be broken by this change.

[Kojoley]

Stop, why we should support the code that uses mpl types and not including mpl? Is not it a bug in their code?

[pdimov]

It is, in my opinion, a bug in their code; and the answer to your question is probably "because code in unmaintained libraries will break and we'll need to fix it." :-)

[Kojoley]

If the only concerns are about Boosts code, I volunteer myself to identify and fix them.

[Kojoley]

I made a list of libraries with grep -ER "bool_\s*<\s*(true|false)\s*>" libs | cut -d / -f 2 | sort | uniq, and run their tests with the PR applied https://travis-ci.com/Kojoley/type_traits_experiment/builds/143657865 and they are fine, except timeouted multiprecision and not related issues in hof/statechart.