No, there's no public documentation, although the thread you posted (which was posted in the old, and now read-only, forum) gives some hints to it being a zip file.
Exporting a csv is inherently slow because it's an inefficient way to store the data, I would instead suggest saving your waveforms as .wfm or .mat files, which have public documentation and result in far smaller files.
2) I am not sure why it's missing from one of your files, would you be able to share a picture of TekScope when you try to save a waveform which leads to this issue?
Yes, under the "Save As" menu, you can select several ways to save the acquisition (tiq, csv, mat, etc), which will save the time domain data as IF or IQ data.
You can also use the SCPI interface to get this data programmatically.
IQ sample data is real-time data, so you cannot capture more than 40 MHz of it in real-time with the RSA306B. The RSA5126B can do up to 160 MHz.
The only RSA that can do 800 MHz is the RSA7100A/B.
If you don't need real-time data, then you can set your span to 800 MHz and then save the trace data as a .specan file (which is really just an XML file that you can open up as a text file and get each point).
Obviously, this will not work well if your signal varies with time since the RSA's you have need to sweep to do an 800 MHz span.
That hints that there might be an error with your file, please try downloading it again and then follow the installation instructions in the release notes.