[MLIR][Transform][Python] Sync derived classes and their wrappers

[rolfmorel]

Updates the derived Op-classes for the main transform ops to have all the arguments, etc, from the auto-generated classes. Additionally updates and adds missing snake_case wrappers for the derived classes which shadow the snake_case wrappers of the auto-generated classes, which were hitherto exposed alongside the derived classes.

[github-actions]

✅ With the latest revision this PR passed the Python code formatter.

[rolfmorel]

Meant to address: #166765

[ftynse]

Can we somehow automate generation of these wrappers? They are marginally more usable, but seem to have high maintenance cost.

[rolfmorel]

Discussed it with @makslevental yesterday. In the case of these transform ops, the derived classes and wrappers really only serve two purposes: add type hints and add _ods_common.get_op_result_or_value around operand arguments. Both of these could in principle be derived from tablegen and be part of the auto-generated classes and wrappers.

From my perspective, the main issue is the mapping from tablegen types, i.e. C++ types, to the type hints. I figure it would be something like

llvm-project/mlir/tools/mlir-tblgen/OpPythonBindingGen.cpp

Lines 492 to 556 in 3719c43

	static std::string getPythonAttrName(mlir::tblgen::Attribute attr) {
	auto storageTypeStr = attr.getStorageType();
	if (storageTypeStr == "::mlir::AffineMapAttr")
	return "AffineMapAttr";
	if (storageTypeStr == "::mlir::ArrayAttr")
	return "ArrayAttr";
	if (storageTypeStr == "::mlir::BoolAttr")
	return "BoolAttr";
	if (storageTypeStr == "::mlir::DenseBoolArrayAttr")
	return "DenseBoolArrayAttr";
	if (storageTypeStr == "::mlir::DenseElementsAttr") {
	llvm::StringSet<> superClasses;
	for (const Record *sc : attr.getDef().getSuperClasses())
	superClasses.insert(sc->getNameInitAsString());
	if (superClasses.contains("FloatElementsAttr") ||
	superClasses.contains("RankedFloatElementsAttr")) {
	return "DenseFPElementsAttr";
	}
	return "DenseElementsAttr";
	}
	if (storageTypeStr == "::mlir::DenseF32ArrayAttr")
	return "DenseF32ArrayAttr";
	if (storageTypeStr == "::mlir::DenseF64ArrayAttr")
	return "DenseF64ArrayAttr";
	if (storageTypeStr == "::mlir::DenseFPElementsAttr")
	return "DenseFPElementsAttr";
	if (storageTypeStr == "::mlir::DenseI16ArrayAttr")
	return "DenseI16ArrayAttr";
	if (storageTypeStr == "::mlir::DenseI32ArrayAttr")
	return "DenseI32ArrayAttr";
	if (storageTypeStr == "::mlir::DenseI64ArrayAttr")
	return "DenseI64ArrayAttr";
	if (storageTypeStr == "::mlir::DenseI8ArrayAttr")
	return "DenseI8ArrayAttr";
	if (storageTypeStr == "::mlir::DenseIntElementsAttr")
	return "DenseIntElementsAttr";
	if (storageTypeStr == "::mlir::DenseResourceElementsAttr")
	return "DenseResourceElementsAttr";
	if (storageTypeStr == "::mlir::DictionaryAttr")
	return "DictAttr";
	if (storageTypeStr == "::mlir::FlatSymbolRefAttr")
	return "FlatSymbolRefAttr";
	if (storageTypeStr == "::mlir::FloatAttr")
	return "FloatAttr";
	if (storageTypeStr == "::mlir::IntegerAttr") {
	if (attr.getAttrDefName().str() == "I1Attr")
	return "BoolAttr";
	return "IntegerAttr";
	}
	if (storageTypeStr == "::mlir::IntegerSetAttr")
	return "IntegerSetAttr";
	if (storageTypeStr == "::mlir::OpaqueAttr")
	return "OpaqueAttr";
	if (storageTypeStr == "::mlir::StridedLayoutAttr")
	return "StridedLayoutAttr";
	if (storageTypeStr == "::mlir::StringAttr")
	return "StringAttr";
	if (storageTypeStr == "::mlir::SymbolRefAttr")
	return "SymbolRefAttr";
	if (storageTypeStr == "::mlir::TypeAttr")
	return "TypeAttr";
	if (storageTypeStr == "::mlir::UnitAttr")
	return "UnitAttr";
	return "Attribute";
	}

When I find some time, I can try implement something like that. If someone has a better idea of how to encode the mapping from MLIR/C++ types to Python type hints, please let me know!