perf: trim body bytes before UTF-8 decoding

[AmisKwok]

Summary / 摘要

Trim request/response body bytes to the panel's preview cap before UTF-8 decoding, so a 512 KB body is no longer decoded in full just to keep the first 32K characters. / 在 UTF-8 解码前按面板预览上限裁剪请求 / 响应体字节，不再为保留前 32K 字符而完整解码 512 KB 的 body。

Changes / 变更

• Add _bytesForPreviewDecode in http_interceptor.dart (shared by both part files): clips the byte prefix to maxChars * 4 + 3 — one UTF-8 char is at most 4 bytes, plus 3 so a multi-byte char is never cut in half. / 在 http_interceptor.dart 新增 _bytesForPreviewDecode（两个 part 共用）：把字节前缀裁剪到 maxChars * 4 + 3 —— UTF-8 单字符最多 4 字节，额外 3 字节确保不会把多字节字符切在中间。
• Response side: _decodeBodyOrHexPreview takes maxChars and decodes only the trimmed prefix; the base64 / hex binary fallback still reads the original bytes, so binary responses behave exactly as before. / 响应侧：_decodeBodyOrHexPreview 新增 maxChars，只解码裁剪后的前缀；base64 / hex 的二进制兜底仍基于原始字节，二进制响应行为完全不变。
• Request side (inspector_http_client.dart): same trimming before utf8.decode for captured request bodies. / 请求侧（inspector_http_client.dart）：请求体解码前同样剪枝。
• Add the InspectorService.maxBodyPreviewBytes getter so interceptors can read the configured preview cap (it used to be private and invisible to them). / 新增 InspectorService.maxBodyPreviewBytes getter，让拦截器能读取配置的预览上限（原来该上限私有、拦截器无法感知）。
• Bump version to 1.13.0 across pubspec.yaml, ios/zero_inspector_kit.podspec, InspectorVersion.value, both READMEs (install snippets + upgrade callout) and CHANGELOG.md; regenerate docs/ via sync-docs.mjs. / 版本升至 1.13.0，同步 pubspec.yaml、ios/zero_inspector_kit.podspec、InspectorVersion.value、两个 README（安装片段与升级提示）与 CHANGELOG.md；并通过 sync-docs.mjs 重新生成 docs/。

Context / 背景

The interceptor buffers up to 512 KB of a body, but NetworkRequest.copyWith(maxBodyBytes:) truncates the preview to 32K characters — so the decode paid for a tail that is discarded immediately (up to ~16x wasted work for ASCII, ~4x for CJK). That decode is synchronous on the UI isolate, and it was the only realistic way the inspector could cost the host app a frame on large responses. / 拦截器最多缓冲 512 KB body，而 NetworkRequest.copyWith(maxBodyBytes:) 会把预览截断到 32K 字符 —— 解码成本全花在随即被丢弃的尾部（纯 ASCII 最多浪费约 16 倍，中文约 4 倍）。该解码同步发生在 UI isolate 上，是检查器在超大响应时唯一可能让宿主 App 掉一帧的地方。

An extra isolate was considered and rejected: Flutter isolates share no memory, so the cross-isolate copy costs as much as (or more than) the decode it saves, and the capture probes — HttpOverrides, debugPrint override, addTimingsCallback, NavigatorObserver — must live in the main isolate anyway. Preview content and length are unchanged by this PR. / 曾评估过另开 isolate，已否决：Flutter isolate 之间不共享内存，跨 isolate 拷贝的成本不低于省下的解码成本，而且采集探针（HttpOverrides、debugPrint 覆写、addTimingsCallback、NavigatorObserver）本身必须在主 isolate。本次改动不改变预览内容与长度。

Checklist / 检查项

• Title follows Conventional Commits / 标题符合约定式提交
• CI checks pass after merge / 合入后 CI 通过

Test plan

• dart format . — 113 files, 0 changed
• flutter analyze — No issues found
• flutter test — 285 passed (incl. version_consistency_test)
• dart run tool/check_release.dart — all version references consistent with 1.13.0
• Manual: fire a response larger than 512 KB and confirm the panel still shows exactly the first 32K characters plus the [… truncated N chars …] marker / 手工：发起一个大于 512 KB 的响应，确认面板仍显示前 32K 字符与 [… truncated N chars …] 提示
• Manual: fire a binary (image / gzip) response and confirm it still renders via the base64 or hex fallback instead of being misread as text / 手工：发起二进制（图片 / gzip）响应，确认仍走 base64 或 hex 兜底，不会被误判成文本

🤖 Generated with Zero Buddy