Those plots show the relationship between mean ChIP tag counts vs. mappability or GC content, when we only look at the bins with input tag counts in certain range.
Input sample is supposed to reflect background, including mappability and GC content. If this is the case, there should be no specific pattern in M|input and GC|input plots because input sample already explains all the information. However, when ChIP or input samples have relatively low sequencing depth, as the ones presented in the vignette, information in input samples is still limited and we observe non-random patterns.
And in these plots, mean ChIP tag counts can be larger than one and it is between 0 and 1 just by chance.