"Line width" isn't a standard term in FFT analysis, so you'd ultimately want to ask SRS what they mean by that. It's a property of the signal being analyzed, not the analysis process itself.
For a real-only FFT, the Nyquist frequency (fs/2) appears at the halfway point in the set of output bins. The upper half contains no unique information, just a mirrored copy of the lower half, so it's discarded. So if you start with a 256-point time record at fs=102.4 kHz, your FFT bins are 400 Hz wide. Bin zero contains the DC component, and the Nyquist frequency at 51.2 kHz appears at bin 128. Your signal should be at zero amplitude by the time that point is reached, so the useful information would be limited to bins 0-127. "Line width" never comes into this picture, but bandwidth does. The contents of a given bin represent not only energy at that bin's center frequency but leakage from the center frequencies of adjacent bins as well... along with all other possible frequencies in the vicinity.
Your window function will determine the usable frequency resolution that you can achieve at a given record width. As you've noticed the window function also has the effect of generating questionable data at the very beginning of the spectrum near DC. Between the effects of the window function and the antialiasing filter, you often need to discard the first and last few bins in real-world FFT applications.
All of this stuff, of course, makes it harder to answer simple questions like, "Can I make noise measurements at 195 mHz?"
You will discover (if you haven't already) that the effective bin width due to the window function may be different for noiselike signals versus CW tones. At the end of the day, the safest thing to do is acquire a long time record, move the data to your PC, and crunch it yourself in MATLAB or Octave. Your only alternative is to become
very familiar with exactly what your SR780 is doing.