Refactor replace and sort

[RAMitchell]

No description provided.

[Copilot]

Pull Request Overview

This PR refactors the replace and sort operations by splitting CUDA-specific GPU code into separate files and modernizing the test implementation. The changes migrate from CUDF-based tests to Arrow-based tests while maintaining backward compatibility for CUDA operations.

Key changes:

• Separates GPU variants into dedicated .cu files for better code organization
• Modernizes test infrastructure from CUDF to Arrow
• Exposes task classes and helper functions in public headers

Reviewed Changes

Copilot reviewed 8 out of 8 changed files in this pull request and generated 3 comments.

Show a summary per file

File	Description
cpp/tests/test_replace.cpp	Migrates from CUDF to Arrow-based test framework
cpp/src/sort.cu	New file containing GPU-specific sort implementation
cpp/src/sort.cpp	Refactored to remove GPU code and expose shared functions
cpp/src/replace.cu	New file containing GPU-specific replace implementation
cpp/src/replace.cpp	Refactored to remove GPU code and update type checking
cpp/include/legate_dataframe/sort.hpp	Exposes task class and helper functions
cpp/include/legate_dataframe/replace.hpp	Exposes task class definition
cpp/include/legate_dataframe/reduction.hpp	Removes unused variant options

---

_{Tip: Customize your code reviews with copilot-instructions.md. Create the file or learn how to get started.}

[Copilot]

Variable name 'null_precendence_with_rank' contains a spelling error. It should be 'null_precedence_with_rank' (precedence, not precendence).

Suggested change

	null_precendence_with_rank,
	auto null_precedence_with_rank = null_precedence;
	null_precedence_with_rank.push_back(cudf::null_order::AFTER);
	auto split_indices = cudf::lower_bound(sorted_table_with_rank.select(keys_idxx),
	global_split_values.select(value_keysx),
	column_order_with_rank,
	null_precedence_with_rank,

[Copilot]

Variable name 'null_precendence_with_rank' contains a spelling error. It should be 'null_precedence_with_rank' (precedence, not precendence).

Suggested change

	null_precendence_with_rank,
	auto null_precedence_with_rank = null_precedence;
	null_precedence_with_rank.push_back(cudf::null_order::AFTER);
	auto split_indices = cudf::lower_bound(sorted_table_with_rank.select(keys_idxx),
	global_split_values.select(value_keysx),
	column_order_with_rank,
	null_precedence_with_rank,

[Copilot]

Variable name 'null_precendence_with_rank' contains a spelling error. It should be 'null_precedence_with_rank' (precedence, not precendence).

Suggested change

	null_precendence_with_rank,
	auto null_precedence_with_rank = null_precedence;
	null_precedence_with_rank.push_back(cudf::null_order::AFTER);
	auto split_indices = cudf::lower_bound(sorted_table_with_rank.select(keys_idxx),
	global_split_values.select(value_keysx),
	column_order_with_rank,
	null_precedence_with_rank,

[seberg]

A lot of code that needed to move here, but looks straight forward. Nice, thanks!