Update the workflow using 1ESPipelineTemplates

…iable shadowing (#2348)

Variable shadowing (reusing value names) is disallowed in ONNX across
the main graph and subgraphs according to the spec
(https://github.com/onnx/onnx/pull/6955/files). This change updates to
the logic to check and raise on such cases. A subsequent PR will
implement #1432 to allow users to fix names explicitly.

Fix a seeming bug in handling of cross-attention in MHA (to be
verified): In MHA fusion, we start with an input graph where attention
is applied to 4D query/key/value, and it is transformed into a MHA op on
3D query/key/value.

In the case of cross-attention (with no rotary-embedding): the fusion
seems to convert just query to 3D, and seems to leave key and value as
4D, which seems wrong.

This PR adds the necessary 4D=>3D conversion for key/value before MHA.

Note: This is a quick fix for the relevant case (that shows up). Other
combinations may be worth checking out separately.

---------

Signed-off-by: Ganesan Ramalingam <grama@microsoft.com>
Co-authored-by: Justin Chu <justinchuby@users.noreply.github.com>

When we reshape from a dynamic shape, the shape can contain zeros. This
change accounts for those cases.

Fix pow.Tensor_Scalar type promotion by accounting different combination
of input dtypes. This change ensures the inputs to Pow is always the
same type for compatibility with downstream tools.

Also
- Added is_floating_point for dtype for convienience. The method naming
follows
https://docs.pytorch.org/docs/stable/generated/torch.is_floating_point.html
- Simplify value str when it is constant.

Fix #2213

Python 3.8 has reached EOL.

Fix #2233

This pull request adds a `__repr__` method to the `MatchResult` class in
`onnxscript/rewriter/_basics.py`. The new method provides a string
representation of the match result, improving debugging and readability.

Take `onnx_ir` as a dependency and expose onnxscript.ir as an alias of
`onnx_ir`.