Numerically stable false alarm probability for Lomb Scargle

[johnh2o2]

Current implementation of fap_baluev is numerically unstable and will return 0 when FAP is small enough. There should be a way to obtain a numerically stable implementation (perhaps using a log(-log(FAP)) transform or something).