Floating point precision is unlikely to be the problem here.
Since you're not trying to coordinate simultaneous use, but just determine which of two sequential sessions came earlier/later, you'll generally be looking at differences in timestamps of minutes, not milliseconds.
Even a single precision float can keep precision to one second for about 33 million seconds, according to the precision table I worked out here, and that's a little more than a year of continuous play. Probably safe unless you're making an idle game that you expect players will keep running unattended 24/7.
You do get inaccuracies from adding floats together, especially if they're very different in magnitude, so don't increment your total playtime using deltaTime every frame — accumulate it in chunks when you save. You'll still lose time to imprecision here, but again your precision needs are so low that it's unlikely to add up to anything impactful.
So no, this doesn't appear to be unsafe from a pure precision point of view. The problems you need to solve are likely to be elsewhere.
For instance, if I load my save on two different devices A and B, then play B offline for two days before syncing it to the cloud. Then play A offline for three days before syncing it to the cloud. Now A's total time played is more than the 2 days I'd previously synched to the cloud, so the server overwrites my B save with A. Is that the desired outcome? Should your game detect and warn the player of this branch in the save history before synching?