fix(cli): change tokenize output to flat {"tokens": [[...]]} format

See merge request apro/mcp/lexsim!2