Add parsetree support for wildcards in simple expressions

[dra27]

This PR first extends the parser to recognise syntax errors specifically related to invalid uses of _ in more places. It then converts these illegal _ to a new parsetree node, moving the error handling for encountering them to the type checker. The value of this change is that ppxes can both transform parsetrees which include the new node, and also use the OCaml parser to parse attribute and extension point payloads which include them.

The names Pexp_hole and Pmod_hole originate from ocaml/merlin#1242 where they were temporarily part of Merlin's AST before being replaced with extensions in ocaml/merlin#1503 in order to remain compatible with ppx infrastructure. Pexp_hole returned in a (highly experimental) feature in oxcaml/oxcaml#3310 which introduced it to the (OxCaml) compiler. Although the feature in OxCaml remains incomplete, we have found the node to be very useful in ppxes, and the parsing was extended in oxcaml/oxcaml#4956 entirely with ppx use in mind. We think the node is a useful addition to the parsetree, even without a language feature that directly requires it. We haven't used Pmod_hole in ppxes, but the intention is that having both Pexp_hole and Pmod_hole in the AST means that Merlin can use it as well (see OxCaml's Merlin changes in oxcaml/oxcaml#6960).

The PR itself is five fairly straightforward commits, which are worth reviewing separately:

• The first two add a parse-errors test which shows the existing fun_expr uses of _ displaying the existing "unexpected wildcard" parsing error, and all the additional simple_expr and module_expr points where a standard "syntax error" is displayed
• The third commit (essentially taking the grammar changes of Parse Pexp_hole as a simple expression oxcaml/oxcaml#4956, without the parsetree node), extends the handling of _ to simple_expr, which can be seen in the parse-errors test output
• The fourth commit then creates the parse tree node for Pexp_hole, removing the hack from the parser and instead converting the error to come directly from the type checker
• The fifth commit does the same for Pmod_hole

There are various subtleties (visible in the reference files) to do with _ and ~/? corner cases. For information, the original work predates LLM-use - Claude's responsibility in this PR was rebasing the work from OxCaml, refactoring the work to be a syntax change, then a parsetree change, and adding the tests to demonstrate the effects.