Using syslib without full suite of primary spectra

[mgerbino]

This issue is related to simonsobs/LAT_MFLike#99. When the user runs with partial set of spectra (e.g., TE/EE only), the code is broken as it attempts to fill in entries in the cl dictionary which should not exist. This is partly addressed by #22 for TE/EE only case in the polangle systematics case, however note that BB/EB/TB only case would not run.

A quick fit is to include checks on entries of requested_cls dict. However, this may slow down the code execution in MCMC mode as checks are repeated at each MCMC step. Any idea?