Use new constant function instead of deleted makeScalar function (#1996)

This seems to be from #1987
(see
https://github.com/openxla/stablehlo/actions/runs/7808382192/job/21298492956).

Not sure how presubmit failed to catch this. I think what happened is
the following:

1. #1987 was uploaded and
passed presubmit.
2. #1983 was submitted, adding
a new use of `makeScalar`.
3. #1987 was submitted,
deleting `makeScalar` and causing tests to fail at HEAD.

What I don't understand is why this sequence of operations is possible
in the first place. I feel like when
#1983 was submitted, the
earlier passing presubmit on
#1987 should have been
invalidated, blocking submit. I guess presubmits only re-run if you add
a commit? So in this case since there was no new commit and no conflict,
no new commit needed to be pushed and the presubmits were not run again.

In the grand scheme of things, I think this is fairly minor, but we may
want to investigate whether we can prevent this somehow by tweaking some
configuration settings or something.

stablehlo/reference/Ops.cpp

@@ -1115,9 +1115,8 @@ Tensor evalCollectiveBroadcastOp(
     return process->rendezvous(*processGroup, channelId, operand)
         .lookup((*processGroup)[0]);
 
-  return evalBroadcastInDimOp(
-      makeScalar(convert(operand.getElementType(), 0.0)), {},
-      operand.getType());
+  return evalBroadcastInDimOp(constant(0.0, operand.getElementType()), {},
+                              operand.getType());
 }
 
 Tensor evalCollectivePermuteOp(