fix: sanitize lone surrogates from stdin to prevent UnicodeEncodeError

[octo-patch]

Fixes #667

Problem

On Windows, piping binary content (e.g. git diff on a repository with binary files) through stdin causes Python's text-mode stdin reader to produce lone surrogate characters via the surrogateescape error handler. These surrogates propagate into the JSON payload sent to the API, and httpx raises:

UnicodeEncodeError: 'utf-8' codec can't encode characters in position 1297-1298: surrogates not allowed

This affects any workflow that pipes content containing non-UTF-8 bytes to sgpt, most commonly on Windows but potentially on other systems with non-UTF-8 locales.

Solution

After collecting all stdin lines into the stdin string, apply a single-pass sanitization:

stdin = stdin.encode("utf-8", errors="replace").decode("utf-8")

This replaces any lone surrogate characters with the Unicode replacement character (U+FFFD) before the string is incorporated into the prompt and eventually serialized to JSON.

Testing

• Added test_stdin_with_surrogate_characters in tests/test_default.py which injects a mock stdin containing a lone surrogate (\udcff) and verifies:
  1. The CLI exits with code 0 (no crash).
  2. The message content passed to the LLM is valid UTF-8 (.encode("utf-8") does not raise).
• All 30 existing tests continue to pass (pytest tests/ --ignore=tests/_integration.py).