[mlir] Remove assert from RegionBranchOpInterface verifier

[Mogball]

Summary

This assert is erroneous because an op implementing
RegionBranchOpInterface can have variadic regions and in some cases
have zero regions, in which case the only possible control flow is
branching from the parent op to itself.

[jpienaar]

Branching to itself == NOP? (On initial read it sounded like infinite loop)

[Mogball]

"Branching back to itself" means that control flow goes from "before the op" to "after the op", so yes it'd be NOP

mlir/lib/Interfaces/ControlFlowInterfaces.cpp

LogicalResult detail::verifyTypesAlongControlFlowEdges(Operation *op) {
   auto inputTypesFromParent = [&](Optional<unsigned> regionNo) -> TypeRange {
     return regionInterface.getSuccessorEntryOperands(regionNo).getTypes();
   };
 
   // Verify types along control flow edges originating from the parent.
   if (failed(verifyTypesAlongAllEdges(op, llvm::None, inputTypesFromParent)))
     return failure();
 
-  // RegionBranchOpInterface should not be implemented by Ops that do not have
-  // attached regions.
-  assert(op->getNumRegions() != 0);
-
   auto areTypesCompatible = [&](TypeRange lhs, TypeRange rhs) {
     if (lhs.size() != rhs.size())
       return false;
     for (auto types : llvm::zip(lhs, rhs)) {
       if (!regionInterface.areTypesCompatible(std::get<0>(types),
                                               std::get<1>(types))) {
         return false;
       }