[AscendNPU-IR][AI] Support skill to visualize precision error debugging

[mmyy1217]

1. Add precision ASCII-visualization tool
2. Add agent skill related to this precision ASCII-visualization tool

[github-actions]

👋 Hi! Thank you for contributing to the TileLang project.

Please remember to run bash format.sh in the root directory of the project to ensure your changes are properly linted and formatted. This will help ensure your contribution passes the format check.

We appreciate you taking this step! Our team will review your contribution, and we look forward to your awesome work!

🚀

[gemini-code-assist]

Code Review

This pull request introduces a precision debugging utility, tilelang.utils.precision_debug.assert_close, which provides detailed mismatch reports and ASCII diff maps. The implementation includes new agent skills and evaluation prompts to support automated debugging. Key feedback includes a critical indentation error and discouraged wildcard import in the utilities initialization, a truncated duplicate line in the documentation, and a suggestion to adjust the stack depth traversal for more accurate caller identification.

[gemini-code-assist]

This change introduces a syntax error (IndentationError) because the following lines (18-20) remain indented but are no longer part of an import grouping. Additionally, wildcard imports (from ... import *) are generally discouraged as they pollute the namespace. It is better to maintain the explicit import list and add the new assert_close utility separately.

Suggested change

	from .npu_utils import *
	from .precision_debug import assert_close as assert_close
	from .precision_debug import assert_close
	from .npu_utils import (

[gemini-code-assist]

This line appears to be a truncated duplicate of the previous line and should be removed.

[gemini-code-assist]

The stack depth traversal here (4 levels) seems too deep for capturing the immediate caller of assert_close. Typically, 2 levels are sufficient: one to skip _get_caller_info and one to skip assert_close itself. Using 4 levels might result in reporting internal framework or test runner frames (like pytest) instead of the user's test function.

Suggested change

	for _ in range(4):
	frame = frame.f_back
	for _ in range(2):