merl-an fails silently on code that hasn't been built

[3Rafal]

Running merl-an on repository that hasn't been built returns results that seems ok, but are wrong.

[3Rafal]

@pitag-ha ,
I did some investigation. It seems that merlin returns response with error, as seen on an example below:

          {
            "type": "config",
            "sub": [],
            "valid": true,
            "message": "No config found for file ba5fe479f7d7f1b2757fc7344a7452b8/default/test/bench.t/bench.ml. Try calling `dune build`."
          },

or

      {
        "class": "error",
        "value": "Destruct not allowed on non-immediate type",
        "notifications": [],
        "timing": {
          "clock": 0,
          "cpu": 0,
          "query": 0,
          "pp": 0,
          "reader": 0,
          "ppx": 0,
          "typer": 0,
          "error": 0
        }
      }

However, they are not checked and reported to the user.

[3Rafal]

Similarly to #33, we could check return classes for benchmarks and report the quantity of each responses. That way, having spike of errors/failures would signal that something is wrong.

[pitag-ha]

The build problem is a global problem, right? I.e. either it's a problem for all commands or for none. So we could run ocamlmerlin errors one time at the beginning of the merl-an execution and look for a "No config found" message. If none is found, we assume everything is ok. What do you think?