Fix closure representation in the Nock backend

[lukaszcz]

• Closes Nockma backend: Adjust closure implementation to match the Nock calling convention #3083
• Closes Nockma backend: reduce the number of things stored in closures #3042

The representation of closures is changed to make it more efficient and compatible with the Nock calling convention.

[paulcadman]

Is it possible to add additional tests that use currying that would have failed with the previous approach?

[lukaszcz]

Is it possible to add additional tests that use currying that would have failed with the previous approach?

The old way of handling closures is correct "internally" -- as long as we're using the old closure calling convention everywhere. The problem is that the old calling convention doesn't match the Nock calling convention, so it would crash if a partially applied function is called externally by Nock. But we're not testing against the actual Nock interpreter right now, so it's hard to add a new test for that.

This is partially tested in anomaCallingConventionTests where applyFun uses the Nock calling convention (as far as I understand it). It's used with partially applied (curried) functions in the tests, so curryClosure is tested there.

[paulcadman]

Is it possible to add additional tests that use currying that would have failed with the previous approach?

The old way of handling closures is correct "internally" -- as long as we're using the old closure calling convention everywhere. The problem is that the old calling convention doesn't match the Nock calling convention, so it would crash if a partially applied function is called externally by Nock. But we're not testing against the actual Nock interpreter right now, so it's hard to add a new test for that.

This is partially tested in anomaCallingConventionTests where applyFun uses the Nock calling convention (as far as I understand it). It's used with partially applied (curried) functions in the tests, so curryClosure is tested there.

This makes sense thanks.