An interpreter for the IR

[nomeata]

this is currently a pretty straight-forward copy of interpret.ml to
the new IR. The test suite ensures that we get identical output with
either Interpret or Desugar+Interpret_ir (and I checked that the
test suite would report the differences if there were any).

This is still uses the CPS style, and I think this is unavoidable if we
want to be able to interpret our code at any stage of the pipeline,
including before and after the await-translation. Also, for code resuse
in Value (which assumes a CPS’ed function type).

Fixes #141.

[nomeata]

Ok, the remaining differences (which are visible as */ok/*.diff-ir.ok files) are due to the fact that the desugaring pass does not handle class identities properly. This anticipates #81.

[crusso]

Looks good to me, thought the re-use of scopes for the initial environments, though convenient, may be confusing - a scope is the output of checking/evaluating a declaration, not a general environment and it's really just a coincidence that we can use them for our initial environments.

Otherwise, looks good to me.

[nomeata]

thought the re-use of scopes for the initial environments, though convenient, may be confusing

It mirrors what we do for the type environment. I don’t mind a different type, maybe even called “env”, in interpret.mli, as long as it does not have the unused fields it has before (or is simply abstract).