Fix Cursor hook stdin decoding for CJK content

[Siddhant-K-code]

Summary

• Read --hook-input stdin as bytes before decoding
• Support UTF-8 plus UTF-16LE/UTF-16BE hook payloads, including BOM and NUL-pattern detection
• Add a Cursor regression with odd-count CJK content encoded as UTF-16LE stdin

Fixes #1317.

Validation

Run from the repo root on this branch:

task fmt
task lint
task test TEST_FILTER=test_cursor_checkpoint_stdin_with_utf16le_odd_cjk_content

Expected outcomes:

• task fmt exits cleanly with no rustfmt diffs.
• task lint exits cleanly with no clippy/lint failures.
• The targeted Cursor test passes.
• The test sends a UTF-16LE JSON hook payload through --hook-input stdin, the checkpoint command does not emit Invalid JSON in hook_input, and the committed CJK line is verified as AI-authored.

Broader regression sweep:

task test TEST_FILTER=cursor

Expected outcome: Cursor preset and integration tests pass, including UTF-8 BOM and UTF-16 stdin decoding paths.

Additional whitespace check:

git diff --check origin/main..siddhant/1317-cursor-windows-cjk-hook-json

Expected outcome: no output and exit code 0.

Docs

No docs change needed. This is an input decoding bug fix for existing Cursor hook behavior and is covered by regression tests.

Notes

Draft PR for focused review and CI validation.