Fix MSVC build for 4.07-4.09

[dra27]

This should fix #11. Although with the library will be build with ocaml/dune#2693, OCaml 4.07-4.09 can't use the resulting stdlib_shims.cmxa file without ocaml/ocaml#9011 which will require at least OCaml 4.10.0, so a workaround is needed here too.

The workaround I've attempted is to duplicate the Pervasives module. This has the immediate effect of allowing the deprecation warning from 4.08.0 to be changed so that the stdlib-shims library is not recursively mentioned. On OCaml 4.07 this has no effect (cf. ocaml/ocaml#2041 in 4.08.0), so OCaml 4.07.0 doesn't gain the deprecation warning.

The purpose of this library is to provide compatibility with Stdlib for OCaml < 4.07.0. In those versions, it's virtually possible to define a module called Pervasives (that's the purpose of ocaml/ocaml#1010), so if a user uses stdlib-shims then they have already opted out of being able to shadow standard library modules, and the inclusion of our own pervasives.cmi is - I think - safe. It's also the case that if you're using this library, you're meant to be avoiding using Pervasives completely.

[ghost]

I get a bad feeling about adding a toplevel Pervasives module. In theory it should be fine but I can imagine it either being confusing for users or triggering some weird error. What about using something properly explicit, such as Temporary_workaround_to_empty_lib_bug_on_msvc?

[dra27]

I debated that too - what I dislike about it is that if two libraries were to use that trick then you'd be unable to link with both of them, and that seemed the more likely of two very unlikely possibilities (the other being some weird kind of errror we've not considered). Given that the raison d'etre of this library is to stop using Pervasives completely, it seems really unlikely that you'd ever try to link with another library that was trying to define Pervasives too (and, the second you do that, you can't build it with OCaml < 4.08.0, so the use of stdlib-shims is already very strange). Perhaps that's an argument for keeping the redefinition in 4.10+ as well.

The fix if someone manages to find a convincing breaking pattern would also be really quick (that's not an argument in favour of either solution, just that if anything does break, we can fix it)

What do you think?

[ghost]

Seems fine to me, let's get this merged and released!