[circle-mlir] Need to support `onnx.Range` transform

[shs-park]

The ReduceSumSquare pass appears to be working now with the related PRs.
The next failure occurs in the onnx.Range operation.

...
loc("onnx_wrapper/StatefulPartitionedCall/conformer/conformer_encoder/range"): error: failed to legalize operation 'onnx.Range'
Conversion done: -1

Originally posted by @shs-park in #16360

[shs-park]

Does the circle support the Range operation?
Does the circle have the Range operation?

[shs-park]

The Circle IR already has the Range node

• https://github.com/Samsung/ONE/blob/master/compiler/luci/lang/include/luci/IR/Nodes/CircleRange.h

[shs-park]

Let's introduce the RangeOp to circle-mlir.

[shs-park]

In the draft, the onnx2circle conversion itself succeeds, but the test fails in the circle-impexp-test stage.
It seems there may be an issue in the import code that reads the shape_signature field of the Range operation.

...
[ 69%] circle-impexp-test: generate Range_F32_R0_1.2.circle
NYI GetTensorType Tensor shape_signature.
Error reading return types
Could not translate function 'main_graph'
Error can't load file /home/shpark/git/ONE/circle-mlir/build/debug/circle-mlir/tools-test/circle-impexp-test/Range_F32_R0_1.circle

[shs-park]

The Range operation produces a dynamic output shape that depends on the input values.
Similar to the StridedSlice's dynamic-shape case, it should be represented as a dynamic shape,
but value tests should likely be skipped.

[hseok-oh]

Range operation uses scalar(rank=0) inputs with float, int32 or int64 datatype, and your draft is satisfying this spec.