Implement basic Parser

[rengolin]

With the token tree in hand, produce an AST from it. If writing it in opaque MLIR is a good idea (see #101), then output opaque MLIR in a recognisable shape.

Acceptance criteria:

• Can parse the existing Verona examples (write more examples if necessary)
• Can dump the AST in text form
• Has tests that show the expected AST from given source
• Test coverage is 100% of what's exposed in existing examples

Not included:

• Support for the full language
• Type inference, validation
• MLIR lowering

[rengolin]

@plietar I think we should reuse, or at least heavily borrow from your prototype.

[davidchisnall]

The prototype uses Pegmatites, which is designed for teaching and for rapid prototyping. We should not aim to reuse any of the Pegmatite code for a production compiler.

[mjp41]

@sylvanc has a prototype parser in a private repo he was working on that might be a useful starting point.

The current parser has some quirks, that I don't think need replicating. It tries to accept a mix of ML like syntax, and only requires parenthesis where necessary

   if true {..} else {..}

versus

   if (true) {..} else {..}

I think we should just enforce parens as it makes the syntax more regular.

[rengolin]

I agree. Fuzzy syntax creates complications for the compiler for no additional benefits to the user. We may want to re-think this later, by adding syntactic sugar to the language (and its own parsing rules), but that has to be a conscious act for real benefit, not premature laziness.

[rengolin]

Superseeded by #111