convert extract and shuffle to mlir

[ftynse]

Both require special-casing to convert attributes.

We cannot yet lower this further because extract and shuffle don't have index sequences, neither they need to after lowering, but we can't detect that due to

[Copilot]

Pull request overview

This PR adds MLIR conversion support for extract and shuffle operations in the Wave compiler. The changes enable these operations to be properly converted to the Wave MLIR dialect by handling their specific attribute requirements.

Changes:

• Added MLIR conversion logic for Extract and Shuffle operations with special attribute handling
• Added imports for Extract and ShuffleOp operations
• Added a new test case demonstrating the conversion of a sum kernel that uses extract and shuffle operations

Reviewed changes

Copilot reviewed 2 out of 2 changed files in this pull request and generated 2 comments.

File	Description
wave_lang/kernel/wave/mlir_converter/water_emitter.py	Adds Extract and Shuffle operation imports, registers ShuffleOp in the operation mapping, and implements custom MLIR conversion logic for both operations
lit_tests/kernel/wave/mlir_converter.py	Adds a new test case mlir_converter_sum that verifies correct MLIR generation for a kernel using extract and shuffle operations

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[Copilot]

The assertion lacks a descriptive message explaining why the offset length must be 1. Consider adding a message like 'Extract operation requires exactly one offset value' to help with debugging.

Suggested change

	assert len(node.offset) == 1
	assert len(node.offset) == 1, "Extract operation requires exactly one offset value"

[Copilot]

The integer type creation ir.IntegerType.get_signless(32) is duplicated. Consider extracting it to a variable like i32_type to improve maintainability and reduce redundancy.

Suggested change

	offset = ir.IntegerAttr.get(
	ir.IntegerType.get_signless(32), node.offset
	)
	width = ir.IntegerAttr.get(
	ir.IntegerType.get_signless(32), node.width
	)
	i32_type = ir.IntegerType.get_signless(32)
	offset = ir.IntegerAttr.get(i32_type, node.offset)
	width = ir.IntegerAttr.get(i32_type, node.width)