Refactor Type_enclosing and make use of Context information

[voodoos]

This PR:

• Extracts the module Context from analysis/locate.ml
• Extracts some of the Type_enclosing logic from query_commands.ml
• Makes Context provide richer infos
• Use that information in Type_enclosing

This fixes the bug highlighted in #1104
Should also fix #864

[trefis]

Actually, can we split this PR in two?
Let's just keep the extraction commits (the first two) that are obviously good and mergeable, and open a separate one for the use of context in type_enclosing.

Thanks!

[voodoos]

Status of this PR

It is still partial, and when context information is not satisfying it fallback to the old behaviour: try to type in the various namespaces until success. Tests need to be more exhaustive but I believe it should give identical or better results than master. For example it does fix #1104

The weird transition from 405 to 406

As we discussed with @trefis and @let-def, something happened between these two versions.
With this PR the tests context.t and let.t, which were added recently as issues, are indeed behaving as expected for ocaml <= 405.
However past version 406 they do not behave as expected any more.

Todo

• Solve the discrepancy between 405 and 406 (may be in another PR)
• Use context information to type partial paths (for now it falls back to previous behaviour)
• Remove some debug prints

[voodoos]

What changed since last comment:

• Added more tests
• Cleaner Type_enclosing signature
• Refactored type_in_env and removed some need for the fallbacks thanks to better context longident analysis (fallback stil present in the code).
• Test cons.t is now correct from version 406
• Record labels are now correctly typed
• Fixed a bug in locate (527f4a7)
• Tests let.t still wrong

[voodoos]

@trefis do you think this is ready for merging ?

It opens the door to more work, but this can be the subject of other PR's:

• Solve the discrepancy between 405 and 406 (may be in another PR)
• cons.t test still wrong from 406.
• when cursor at (M.|A 3), the enclosing node returned is const 3, thus
  breaking the context inference (but the fallback catch that for now...)

These are not regressions but already present issues.
The last one is the one that prevent us for having a fully context-based type inference. (For now the old fallback behaviour is still in place.)

[trefis]

One last request before this can be merged: there seems to be a lot of whitespace changes in areas of the code / comments that you otherwise do not touch.
Could these be reverted?

[trefis]

Will merge once CI comes back green.