[calyx] fix calyx canonicalization.

[linuxlonelyeagle]

See issue.
Honestly, I saw this bug a few months ago and I think I fixed it, just like this PR did. But I don't think it was enough.

test.mlir:2:3: note: see current operation: 
"calyx.component"() ({
^bb0(%arg0: i1, %arg1: i1, %arg2: i1, %arg3: i1):
  %0:6 = "calyx.register"() {sym_name = "r"} : () -> (i1, i1, i1, i1, i1, i1)
  "calyx.wires"() ({
  ^bb0:
  }) : () -> ()
  "calyx.control"() ({
  ^bb0:
  }) : () -> ()
}) {function_type = (i1, i1, i1, i1) -> (), portAttributes = [{go}, {clk}, {reset}, {done}], portDirections = -8 : i4, portNames = ["go", "clk", "reset", "done"], sym_name = "main"} : () -> ()

The problem is as I said in the issue, solved he still reports an error because now calyx.control is empty after -canonicalize.
So a few months ago I tried to include stricter restrictions, but I didn't have too good an idea.
Now my idea, when checking calyx.control, that op must have calyx.enable,calyx.invoke alive fsm.machine present. but it also causes other problems.
On the one hand, I want to discuss it, and on the other hand, I think this PR makes sense.There is no test file because there are no examples of errors reported after -canonicalise in the existing test file.

[rachitnigam]

Thanks for creating this @linuxlonelyeagle! I think we should add a text with the three cases you mentioned so that future PRs don't break it. I'm also wondering if we should start canonicalizing away empty control programs (if with empty bodies etc.) so passes can make stronger assumptions about the shape of programs.

[cgyurgyik]

A few smaller comments, but otherwise LGTM.

[cgyurgyik]

I'm also wondering if we should start canonicalizing away empty control programs (if with empty bodies etc.) so passes can make stronger assumptions about the shape of programs.

+1

[cgyurgyik]

LGTM, I see this as an incremental improvement. Thanks!

[linuxlonelyeagle]

Can you help me merge codes? Thanks!