Handle small tactical/openings NPZ batches

[lukifer23]

Summary

• clamp tactical and openings sampling to the available curriculum data, enabling replacement sampling only when needed
• skip empty tactical/openings shards to avoid malformed outputs
• add regression tests that cover sampling from undersized tactical and openings NPZ files

Testing

• pytest tests/test_data_manager_batches.py

---

https://chatgpt.com/codex/tasks/task_e_68e6f991ddcc83239164246fad36d7cc

[Copilot]

Pull Request Overview

This PR addresses the handling of small tactical and openings NPZ batch files to prevent runtime errors when sampling more data than available. The changes enable robust sampling from undersized curriculum data sources and skip empty data files.

Key changes include:

• Clamping tactical and openings sampling to the available curriculum data size
• Enabling replacement sampling only when the batch size exceeds the available data
• Adding empty data validation to skip malformed outputs
• Including regression tests for undersized NPZ file scenarios

Reviewed Changes

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

File	Description
tests/test_data_manager_batches.py	Adds regression tests that verify batch sampling from undersized tactical and openings NPZ files
azchess/data_manager.py	Updates tactical and openings batch methods to handle small files with proper size validation and replacement sampling

---

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

[Copilot]

[nitpick] The variable draw_size is misleading since it represents the sample size when replacement is disabled, not the size being drawn. Consider renaming to sample_size_no_replace or clamped_size for clarity.

Suggested change

	draw_size = min(batch_size, total_positions)
	replace = batch_size > total_positions
	target_size = batch_size if replace else draw_size
	sample_size_no_replace = min(batch_size, total_positions)
	replace = batch_size > total_positions
	target_size = batch_size if replace else sample_size_no_replace

[Copilot]

[nitpick] The variable draw_size is misleading since it represents the sample size when replacement is disabled, not the size being drawn. Consider renaming to sample_size_no_replace or clamped_size for clarity.

Suggested change

	draw_size = min(batch_size, total_positions)
	replace = batch_size > total_positions
	target_size = batch_size if replace else draw_size
	clamped_size = min(batch_size, total_positions)
	replace = batch_size > total_positions
	target_size = batch_size if replace else clamped_size