Fix BroadcastInDimOps linalg lowering with compatible dim=1

[FruitClover]

For the case of simple broadcast check that we don't have to extend dimentions, so skip convertion to BroadcastOp when only one of them is 1.

Error before:

test.mlir:5:8: error: 'linalg.broadcast' op input dim 1 should match init dim 2. input: 1, init: 5
  %0 = stablehlo.broadcast_in_dim %arg, dims = [1, 2, 3] : (tensor<3x1x1xf32>) -> tensor<1x3x5x7xf32>
       ^
test.mlir:5:8: note: see current operation:
%1 = "linalg.broadcast"(%arg0, %0) <{dimensions = array<i64: 0>}> ({
^bb0(%arg1: f32, %arg2: f32):
  "linalg.yield"(%arg1) : (f32) -> ()
}) : (tensor<3x1x1xf32>, tensor<1x3x5x7xf32>) -> tensor<1x3x5x7xf32>

[google-cla]

Thanks for your pull request! It looks like this may be your first contribution to a Google open source project. Before we can look at your pull request, you'll need to sign a Contributor License Agreement (CLA).

View this failed invocation of the CLA check for more information.

For the most up to date status, view the checks section at the bottom of the pull request.

[GleasonK]

To clarify the difference between simple broadcast / not simple broadcast, it's that extending rank is OK, but for dims where rank is not extended the dim sizes must match?

simple: tensor<3xf32> -> tensor<1x2x3xf32>
not simple: tensor<3x1xf32> -> tensor<1x3x3xf32>

[FruitClover]

To clarify the difference between simple broadcast / not simple broadcast, it's that extending rank is OK, but for dims where rank is not extended the dim sizes must match?

simple: tensor<3xf32> -> tensor<1x2x3xf32>
not simple: tensor<3x1xf32> -> tensor<1x3x3xf32>

basically yes, as i understood from from BroadcastOp semantics