When dealing with classes, the typedtree contains a number of ways to
refer to the type. For example, compiling this:

```ocaml
class type u = object end
```

results in a typedtree containing:

```
Typedtree.Tsig_class_type
  [{Typedtree.ci_virt = Asttypes.Concrete; ci_params = [];
    ci_id_name = {Asttypes.txt = ...; loc = ...}; ci_id_class = u/13[14];
    ci_id_class_type = u/12[14]; ci_id_object = u/11[14];
    ci_id_typehash = #u/10[14];
...
```

these ids may be referenced later in the file, for example:

```ocaml
val g : #u
```

results in:

```
Typedtree.sig_desc =
     Typedtree.Tsig_value
      {Typedtree.val_id = g/14;
       val_name =
        {Asttypes.txt = "g";
         loc = ...
       val_desc =
        {Typedtree.ctyp_desc =
          Typedtree.Ttyp_class (Path.Pident #u/10[17],
           {Asttypes.txt = Longident.Lident "u";
...
```

so the Path.Pident is referring to `#u/10` in this case. When odoc
reads this these idents are turned into odoc `Identifiers`.
Concretely, we add a map from each of the four idents `u/13`, `u/12`,
`u/11` and `#u/10` to the same identifier, so we end up dropping the
explicit '#'.

In contrast, when the path is within a module `M`, and we reference the
class type like `val g : #M.u`, we get this path instead:
`Path.Pdot (Path.Pident M, "#u")`
Internally, odoc turns this into the identifier
`Doc(Identifier M, "#u")` - ie, we don't currently drop the hash. This
patch addresses this by explicitly removing the hash from paths with a
Dot constructor.