fix: don't reject UTF-8 text files dense in CJK as binary

[he-yufeng]

Problem

OHEditor.view() (and str_replace, insert, etc.) raises FileValidationError: File appears to be binary… when operating on perfectly valid UTF-8 markdown files that contain dense CJK (Chinese/Japanese/Korean) characters.

Reported in OpenHands/OpenHands#13517.

Minimal repro:

from openhands_aci.editor.editor import OHEditor
import tempfile

with tempfile.NamedTemporaryFile(suffix='.md', delete=False, mode='w', encoding='utf-8') as f:
    f.write('中文测试内容。' * 50)   # 1050 bytes of valid UTF-8
    path = f.name

OHEditor()(command='view', path=path)
# => FileValidationError: File appears to be binary and this file type cannot be read or edited by this tool.

Root cause

validate_file defers the text/binary decision to binaryornot.check.is_binary, which uses chardet internally. Once a UTF-8 text file crosses ~1 KB with dense multi-byte characters (no ASCII dilution), chardet's byte-ratio heuristics cross the binary threshold and is_binary returns True — even though the bytes decode cleanly as UTF-8.

Dense CJK bytes	is_binary() result
< ~1000	False (correct)
>= ~1050	True (false positive)

This affects anyone working with non-trivial amounts of CJK text (also reproduces with dense Cyrillic, Arabic, etc.).

Fix

Before raising FileValidationError, do a second-chance decode using the existing EncodingManager (which uses charset_normalizer, not chardet, and handles CJK correctly). If we can decode a 64 KB sample with the detected encoding, the file is text.

• Genuinely binary files (null bytes, random bytes) still fail both checks and are rejected.
• No new dependencies — charset_normalizer is already in use via EncodingManager.
• Preserves the fast path: is_binary returning False still accepts immediately; the second-chance decode only runs on files is_binary flagged.

Testing

New regression tests in tests/unit/test_file_validation.py:

• test_validate_dense_cjk_text_file — 1050-byte dense Chinese markdown, asserts is_binary still returns True so we're actually exercising the fix, then asserts validate_file accepts the file.
• test_validate_bom_utf16_text_file — UTF-16 encoded Japanese/Chinese text.

All existing tests that were passing on main still pass. Two tests (test_validate_binary_file, test_validate_image_file) fail locally on both main and this branch due to a binaryornot/Python-3.12 compat issue (NameError: name 'unicode' is not defined inside binaryornot/helpers.py:106) — unrelated to this PR.

Fixes OpenHands/OpenHands#13517