The test samples are from Shain et al. 2015, Nature Genetics. Since these sequences are protected patient information the BAM files were submitted to dbGaP; there is a few months' delay before they appear online. However, those weren't ideal samples for testing copy number calling anyway -- desmoplastic melanoma genomes are dominated by somatic SNVs, not large-scale copy number alterations.
A better dataset for testing variant callers, both SNV and CNV, has become available recently: "An open access pilot freely sharing cancer genomic data from participants in Texas". I recommend running your benchmarks with these samples instead so that you can freely share your complete analysis. CNVkit has changed significantly since the version I benchmarked in the paper, so in any case you'll need to re-run the latest version of each caller (including CNVkit) to get representative results.