Fix compact context length parsing in preprocess

[mySebbe]

Summary

• parse compact result lengths such as 8k/8K into numeric token counts in preprocess.py
• keep numeric lengths such as 0, 8000, and 16000 working as before
• move preprocessing into small functions so the parser can be unit tested without running the whole script on import
• add regression coverage for the compact-length case reported in length parsing error in preprocess.py #8

Verification

• python -m unittest discover -s tests -v
• python -m compileall -q gsm-infinite tests
• reproduced the original failure with length: 8k, then verified python gsm-infinite/preprocess.py processes the same result file successfully

Fixes #8