[RefC] [Cleanup] Pattern matching generates simpler code.

[seagull-kamome]

Description

The jump tables generated by pattern matching seem completely unnecessary. A simpler conditional branch code would reduce malloc/free costs. It also seems to me that debugging would be easier.

Should this change go in the CHANGELOG?

• If this is a fix, user-facing change, a compiler change, or a new paper
  implementation, I have updated CHANGELOG_NEXT.md (and potentially also
  CONTRIBUTORS.md).

[seagull-kamome]

My next plan is to eliminate special constructions and this PR is a preparation for that.

[andrevidela]

Anything that deletes looks good to me, any objectsion?

[mattpolzin]

any objections

only that I know dependent types make case statements deceivingly tricky so (without knowing whether this is a valid concern) I wonder if there are certain dependent case patterns that aren’t fully covered by test cases that would stop working.

On the plus side, there are already certain limitations with the RefC backend so I’m not totally opposed to merging and keeping this in mind if anyone reports new problems that sound like they could be related to this change.

[seagull-kamome]

I added a test in the previous commit.
This test should be placed in allbackends, but for now it is in tests/refc because of the issues in Issue #3161 and #3158.

The biggest problem in the old code is malloc, which should be completely unnecessary. This piles up a lot of small costs that cannot GC when recursion occurs inside the pattern match.

[andrevidela]

I think we're good to go here then