Simple compressed files such as the fastq.gz archive also in that directory @
will upload and automatically uncompress - I just tested that one - it contains a single fastq which Galaxy can handle as expected.
Tarball archives (.tgz) like the one you report cannot be handled (at present!) where they contain multiple files (in this case there are two files, one called "[...]reject.fastq"). The new dataset collections being added to Galaxy may allow this to be fixed in future but for now, one possible but ugly way around this is to grab the .tgz file, uncompress it and upload the bits you want separately - sorry this is not convenient but it will at least work so you can get on with your research?