Support (module definition ...) and (module instance ...) in WAST spec tests

[stevenfontanella]

• Add support for (module definition ...) and (module instance ...) in WAST spec tests
  • Syntax reference
  • These statements are skipped by the WAST splitter, since wasm-opt can't understand them. In the future, we need some extra work to handle explicit module instantiations correctly here.
  • This fixes the memory.wast spec test. The remaining 3 that currently depend on this syntax still break for other reasons (mentioned in shared.py)
• Remove extra unneeded call to run_spec_test

[tlively]

A few initial comments while this is still under construction.

[stevenfontanella]

(Sorry for the force-push, I already made these changes locally before your comments)

[tlively]

Let's move the lexer reset to back before the (module from the caller to here.

[tlively]

Even though they both start with (module, modules and instantiations are parsed very differently and behave very differently, so I think it would make sense to treat them as different things rather than bundling them together like this. That will simplify command below by removing the need for the std::visit variant conversion.

[stevenfontanella]

I mostly agree, but I chose to bundle them together since we must attempt to parse (module instance ...) before attempting (module name? ...), otherwise we'll try to parse the word instance as a name and fail rather than backtracking. Does it make sense to bundle them to make the ordering clear? I agree it's a little unfortunate that the module / module instance parsers need to be aware of each other a little bit but I don't see a way around it.

The only alternative I see is to remove the ordering dependency by letting the module (not instance) parser backtrack if it sees the word instance. That lets the ordering work either way, but that means that both parsers still need to know about each other in which case it makes sense to me to bundle them in the same function.