1) If CNVkit misidentifies the gender of a control sample, I'd usually recommend just not including that sample in the reference pool. The reference samples you use should ideally have a clear signal on all chromosomes, including the sex chromosomes.

Otherwise, you can use two separate reference profiles, one male and one female, then choose the profile to match each tumor sample. Usually that's not necessary, but it's possible to do.

2) You don't need to specify the gender of your tumor samples in fix (it only calculates log2 ratios, no adjustment for sex chromosomes), but you have the option of doing so in call where the log2 ratios are converted into integer copy numbers.