Basically you don't. While it is convention (afaik) to upload the raw data as they come from demultiplexing, the actual uploaded data is what the authors well...uploaded, and this in theory can be anything. There is no bullet-proof way to know beside emailing them.
Though, trimming would usually result in unequal read lengths throughout the files (adapter-containing reads get trimmed, others remain untrimmed) so this is something you can check. I mean in the end it does not really matter, does it? If you want to use the public dataset you are after then you have to use what is provided, and a good QC should always start with something like fastqc to assess whether trimming for adapters or quality was necessary, so this you anyway have to do, regardless how the data have been treated by the uploader before.