ci(encoding): reject invalid UTF-8 and mojibake regressions

[WHUTcjh-2024]

ci(encoding): reject invalid UTF-8 and mojibake regressions

Description | 描述

为受版本控制的文本文件增加可测试的编码守卫，防止 PowerShell/旧代码页链路产生的静默乱码进入仓库。

• 使用 fatal UTF-8 decoder 拒绝非 UTF-8 字节；
• 检测 U+FFFD 替换字符；
• 要求 PowerShell .ps1 使用 UTF-8 BOM，覆盖 issue 中的 L1 防线；
• 精确识别 Windows-1252 和中文旧代码页的常见乱码特征；
• 仅扫描 git ls-files 返回的已跟踪文本文件，不误扫二进制资产和本地产物；
• GitHub Actions 输出文件、行、列注解，可直接定位问题。

检测器允许正常中文、日文、重音拉丁字符和 emoji，避免“禁止所有非 ASCII”造成的误报。

Related Issue | 关联 Issue

Closes #387

Change Type | 修改类型

• Bug fix | Bug 修复
• New feature | 新功能
• Documentation update | 文档更新
• Code optimization | 代码优化

Self-test Checklist | 自测清单

• Verified locally | 本地验证通过
• No existing features affected | 无影响现有功能

Additional Notes | 其他说明

本地验证：

node --test scripts/check-text-encoding.test.mjs
# 8 passed

node scripts/check-text-encoding.mjs
# Encoding check passed (162 tracked text files).

vitest run
# 4 files, 67 tests passed (Node 24.14.0)

npm run build
# passed (Node 24.14.0)

该 PR 仅增加编码检测与 CI 接入，不包含适配器、SDK 或其他功能改动。

[Maxwell-Code07]

Thank you for your attention！ We will conduct a review and respond to you as soon as possible.

[YOMXXX]

Review note after local verification:

The encoding guard itself looks useful and the local verification passed:

• node --test scripts/check-text-encoding.test.mjs (8/8 passed)
• node scripts/check-text-encoding.mjs (162 tracked text files passed)
• COREPACK_ENABLE_AUTO_PIN=0 pnpm test (67/67 passed)
• COREPACK_ENABLE_AUTO_PIN=0 pnpm build
• git diff --check

I would not approve this branch as-is because the PR diff includes unrelated churn outside the encoding guard:

• package.json changes the package version, changes the files entry from src/ to src, and adds zod; these are unrelated to encoding detection.
• .github/workflows/pr-ci.yml appears to have broad line-ending / formatting churn, making the functional CI change harder to review.

Recommended cleanup: keep only the encoding scripts, the two package scripts (check:encoding, test:encoding), and the minimal CI job/step wiring. Removing unrelated package/line-ending churn should make this approve-ready.

[WHUTcjh-2024]

OK, i will see it

[WHUTcjh-2024]

Thanks for the review and verification.

I cleaned this up as requested and force-updated the branch to a single clean commit. The PR diff now keeps only:

• the two encoding scripts
• the two package scripts: check:encoding and test:encoding
• the minimal Encoding Guard CI job wiring

I removed the unrelated package/workflow churn from the final PR diff; there are no package version, files, dependency, or broad line-ending changes left.

Verified locally:

• node --test scripts/check-text-encoding.test.mjs (8/8 passed)
• node scripts/check-text-encoding.mjs (162 tracked text files passed)
• npm run test:encoding
• npm run check:encoding
• local build steps via tsdown and the three script tsc builds
• git diff --check