fix(cli): harden UTF-8 terminal input handling

[cmddirexit]

Summary

• centralize UTF-8-aware reads in the shared CLI console input helper
• temporarily enable termios.IUTF8 while reading UTF-8 input from a TTY
• restore the original terminal attributes after input completes or is interrupted
• keep the chat REPL alive when malformed terminal bytes still raise UnicodeDecodeError
• protect both chat messages and ask_user prompts while leaving non-TTY, non-UTF-8, and unsupported platforms unchanged

Root cause

In canonical mode with IUTF8 disabled, erasing a multibyte character may remove only its final byte. For example, typing the three-byte UTF-8 character 能 and pressing Backspace can leave the incomplete byte sequence e8 83. A strict UTF-8 stdin decoder then raises UnicodeDecodeError, terminating the chat REPL.

Reproduction

1. Run stty -iutf8.
2. Start deeptutor chat with strict UTF-8 stdin decoding.
3. Type a Chinese character, press Backspace, and submit the line.

Tests

• added a POSIX pseudo-terminal regression test that starts with IUTF8 disabled, enters 能, presses Backspace, and verifies both successful input and terminal-mode restoration
• added a REPL test that injects UnicodeDecodeError and verifies the session remains usable
• python -m pytest -q tests/cli (41 passed)
• python -m ruff check deeptutor_cli/chat.py deeptutor_cli/common.py tests/cli/test_chat_cli.py tests/cli/test_chat_terminal.py
• python -m ruff format --check deeptutor_cli/chat.py deeptutor_cli/common.py tests/cli/test_chat_cli.py tests/cli/test_chat_terminal.py

[pancacake]

Thanks for your contribution!