fix(llm): truncate token hints to whole amounts

[spalen0]

Summary

• truncate token-decimal amount hints to whole tokens for values of at least one token
• retain one truncated decimal for sub-token values from 0.1 to under 1, and omit smaller hints
• cover the reported JANE emissions example and sub-token behavior with regression tests

Testing

• /srv/monitoring/.venv/bin/python -m pytest -q (799 passed, 6 skipped)