inference scripts crash on Windows when output is redirected or piped

Describe the bug
The inference scripts print check, cross and warning characters. On Windows, when stdout is redirected or piped (CI, IDE run panes, Git Bash, > file), Python encodes output with the locale code page, usually cp1252, and those characters raise UnicodeEncodeError. The error handler then raises the same exception again while trying to report it. A real console window is not affected, since Python 3.6 and later write to it as UTF-8 (PEP 528).

To Reproduce

1. On Windows, run py inference/validate_labels.py > out.txt.
2. It fails on the first status line with UnicodeEncodeError: 'charmap' codec can't encode character '✓'.

Expected behavior
The scripts run regardless of how output is captured.

Additional context
Setting PYTHONIOENCODING=utf-8 or PYTHONUTF8=1 works around it, which confirms the cause. A code fix would be ASCII status markers, or sys.stdout.reconfigure(encoding="utf-8") at startup. PR #177 replaces some of these in validate_labels.py but not the warning lines. generate_balanced_data.py:218-219 has the same issue. Logger calls in label_generator.py, train_model.py and test_model.py print "--- Logging error ---" tracebacks instead of crashing.

Found while writing the setup documentation in #138 / #178.

Run py inference/validate_labels.py > out.txt on Windows to reproduce the encoding failure. Inspect inference/validate_labels.py, generate_balanced_data.py:218-219, and the logging paths in label_generator.py, train_model.py, and test_model.py. Done means captured or piped output runs without UnicodeEncodeError or logging-error tracebacks.