64-bit float allows more than +/-6165dBFS.
https://www.gearslutz.com/board/showpost.php?p=10582565&postcount=168
http://forum.cakewalk.com/FindPost/2325892
32-bit float allows more than +/-770dBFS.
https://forum.cockos.com/showthread.php?t=215307
https://www.audiosciencereview.com/...attenuation-before-power-amp.5844/post-131431
https://www.audiosciencereview.com/...es-dsd-sound-better-than-pcm.5700/post-135477
Merging and Exasound don't design DAC chips, and DAC chips only understand 24 or 32-bit fixed point data (up to -144.4944 or -192.6592dBFS and zero tolerance above 0dBFS). So bragging about 64-bit float volume control outside of the DAC chip is meaningless at all. The main benefit of floating point processing is avoid clipping above 0dBFS and 32-bit float can do this already.
Therefore regardless of hardware or software digital volume control, the limitation in digital domain will be 24 or 32-bit fixed point. However, all DACs ultimately output analog signal, so real world limitation will be the analog noise floor of the DAC rather than resolution of digital data.
You may benefit from analog volume control if your headphone amp's analog noise floor, measured in analog domain like dBu or dBV, is significantly lower than your DAC's. JDS Labs for example says the Atom amp has -114 dBu noise.
https://www.jdslabs.com/products/190/atom-amp/
After all, it is an unhealthy gain stage if you constantly need to reduce 70dB, it has nothing to do with analog vs digital at all. If you are not planning to replace the headphone amp and worry about channel imbalance, reduce 30dB on the playback software then do the rest on the headphone amp.