Disable smart elaboration of negation in guards

Summary:
As title, do not narrow types under negations. The issue is that, currently, `not (V == a)` is refined identically to `(V == a)`, as the negation is not "propagated".

Occurrence typing is still able to soundly refine types under negations, this is thus only a limitation for functions where occ typing is disabled.

Reviewed By: ilya-klyuchnikov

Differential Revision: D58184266

fbshipit-source-id: 4f71dd99971a7ce0dd73e61b0ebda435952fa4a1

eqwalizer/src/main/scala/com/whatsapp/eqwalizer/tc/ElabGuard.scala

@@ -264,7 +264,10 @@ final class ElabGuard(pipelineContext: PipelineContext) {
  val TestUnOp(op, arg) = unOp
  op match {
  case "not" =>
- (booleanType, elabTestT(arg, booleanType, env))
+ arg match {
+ case TestVar(_) => (booleanType, elabTestT(arg, booleanType, env))
+ case _ => (booleanType, env)
+ }
  case "bnot" | "+" | "-" =>
  (NumberType, elabTestT(arg, NumberType, env))
  case _ =>