Fix token_max_len bug that makes it always zero

[bact]

What does this changes

Fix safe mode algorithm that choose the cut-off position.

What was wrong

token_max_len = 0 was inside for loop, makes it always reset to zero.

This make the cut-off position may not be the one that covers longest token.

How this fixes it

Move token_max_len = 0 outside of for loop.

Your checklist for this pull request

🚨Please review the guidelines for contributing to this repository.

• Passed code styles and structures
• Passed code linting checks and unit test

[coveralls]

Coverage decreased (-0.03%) to 95.934% when pulling ca25598 on bact:fix-newmm-safemode-token_max_len into d51d5a5 on PyThaiNLP:dev.