Spurious inconsistent assumptions

[trefis]

Some time ago we removed the without_cmi thingy, which we did not need anymore with the new version of -short-paths (that is still waiting to be upstreamed).

However recently (i.e. 4.08), other parts of the compiler have been trying to load arbitrary .cmi files, that do not go through the usual "give me the best path" code, here are two I found today: Printtyp.rewrite_double_underscore_paths, Printtyp.Namespace.lookup.
There might be others.

This draft PR adds a test for the first one, I still have to add a test for the second.

As for the fix, the simplest one would be to reintroduce the without_cmi nonsense that the compiler is using.
Was there a reason to remove it other than "we don't need it"? I have the vague feeling there might have been a reason, but I can't remember.

[trefis]

There were issues with without_cmis in the past, but they seem to have been fixed upstream by ocaml/ocaml#1223 .
So I reintroduced it in merlin, which fixes the issues (as shown by the test).

This is ready for review.

[trefis]

I'm reasonably confident this is not going to cause any issue.
So I'm merging now, feel free to complain when you come back from vacation.

[trefis]

Note: this broke the increasing of verbosity when printing expressions types; we sometimes do need to load cmis.
Right now in these situations we end up with this:

Not_found
Raised at file \"map.ml\", line 135, characters 10-25
Called from file \"src/ocaml/typing/408/env.ml\", line 1115, characters 6-28
Called from file \"src/analysis/type_utils.ml\", line 176, characters 17-39
Called from file \"src/ocaml/utils/misc.ml\", line 47, characters 8-15
Re-raised at file \"src/ocaml/utils/misc.ml\", line 59, characters 10-24
Called from file \"src/ocaml/utils/misc.ml\", line 73, characters 10-14
Re-raised at file \"src/ocaml/utils/misc.ml\", line 75, characters 38-45
Called from file \"src/ocaml/typing/408/env.ml\", line 645, characters 10-105
Called from file \"src/utils/std.ml\", line 679, characters 8-12
Re-raised at file \"src/utils/std.ml\", line 681, characters 30-39
Called from file \"src/frontend/query_commands.ml\", line 342, characters 12-132
Called from file \"list.ml\", line 96, characters 20-25
Called from file \"src/utils/local_store.ml\", line 29, characters 8-12
Re-raised at file \"src/utils/local_store.ml\", line 37, characters 4-15
Called from file \"src/kernel/mocaml.ml\", line 40, characters 8-38
Re-raised at file \"src/kernel/mocaml.ml\", line 48, characters 4-15
Called from file \"src/frontend/ocamlmerlin/new/new_commands.ml\", line 65, characters 15-53
Called from file \"src/utils/std.ml\", line 679, characters 8-12
Re-raised at file \"src/utils/std.ml\", line 681, characters 30-39
Called from file \"src/ocaml/utils/misc.ml\", line 47, characters 8-15
Re-raised at file \"src/ocaml/utils/misc.ml\", line 59, characters 10-24
Called from file \"src/frontend/ocamlmerlin/new/new_merlin.ml\", line 99, characters 18-54

Obviously at that point it is fine to load cmis, unfortunately we're already under a wrap_printing_env.
So we should either turn cmi loading back on (which Env.with_cmis doesn't actually do), or push down the call to wrap_printing_env.

[trefis]

More generally: type_utils.ml should be looked at in details now that we've reintroduced without_cmi.