The test updates between the following two PRs:
ocaml#1329
ocaml#1397

There seems to have been an error regression in [locate] between those
two commits. That can be seen both in the category_data.json and in the
full_responses.json.

real    15m27,573s
user    3m5,058s
sys     0m20,955s

This is the change right after the 4.13 bump, i.e. PR
ocaml#1397

There seems to have been an error regression in occurrences (see
category_data.t):
Sometimes it returns an Empty return now, where before it returned a value.

Also, [complete-prefix] has changed (see full_response.t): Sometimes it
adds more options now, e.g. ["name": "acosh"] and ["name": "asinh"]
somewhere.

Also, type aliases now seem to be handled differently sometimes
(see full_response.t): e.g. [Path.t] -> [path];
[P.Contents.value] -> [contents]; [inode_tree] -> [proof_inode] etc

Also, the Typedtree bump is of course reflected in Merlin, e.g.
- [('a, 'b, 'c, 'd, 'e, 'f) Camlinternal
FormatBasics.format6] is now [('a, 'b, 'c, 'd, 'e, 'f) format]

Also, this commit contains some noise that shouldn't be there, because
compiler specific messages weren't filtered out yet.

This is the change between right after 4.13 bump and right before 4.14
bump, i.e. between
ocaml#1397
ocaml#1415

[locate] has changed in two ways: An error regression was fixed (before
[Return Msg], now [Return Other]); there's a change in behavior, where
now it sometimes points to a different place than before.