Dear Elham,
Yes, your model has no states in the rotating frame; in your case, you only need to azimuth-average the matrices without first applying MBC (applying MBC would do no harm as it would have no influence at all).
The very small numbers remaining are the result of numerical round-off in the numerical solution and output. Using FAST compiled in double precision and/or outputting matrices in higher precision would likely reduce the values even further, but my guess is they would still not be exactly zero unless the linearization was formed analytically and all digits of precision were kept in the output. However, for all practical purposes, the values are zero (7 orders of magnitude less than the most dominant term) and you can safely eliminate the generator azimuth angle as a state.
Best regards,