Fix multi-byte Unicode input (CJK/emoji)

[keyuchen21]

Summary

• Fix garbled input when typing Chinese/Japanese/Korean characters or emoji in the chatbox — _raw_read_key() was reading one byte at a time and immediately decoding, corrupting multi-byte UTF-8 sequences into replacement characters (������)
• Fix display width calculations in the input editor — CJK characters occupy 2 terminal columns but were counted as 1, causing cursor misalignment and broken box borders
• _raw_read_available() (used for escape sequences and paste) now accumulates raw bytes before decoding, fixing pasted CJK text

Test plan

• Type Chinese characters (你好世界) — should appear correctly, not as replacement chars
• Type Japanese (こんにちは) and Korean (안녕하세요) — same
• Paste multi-line Chinese text — should insert correctly
• Verify cursor positioning is correct when moving left/right through CJK text
• Verify the input box border stays aligned with wide characters
• Verify normal ASCII input still works as before
• Test emoji input (😀🎉) — should appear correctly

[keyuchen21]

[keyuchen21]

Verification

Tested manually — Chinese input now works correctly:

Before fix: typing "你好" produced "������" (each UTF-8 byte decoded separately as replacement characters)

After fix: "你好" displays correctly in the input box with proper cursor positioning and box alignment

Test results:

• ✅ Chinese input (你好世界) renders correctly
• ✅ Input box borders stay aligned with wide characters
• ✅ Model responds to Chinese input properly
• ✅ ASCII input continues to work as before