> However I would not expect it to then give a clear linear link
No, that's also consistent with healthy user bias: drinking soda and healthy attitudes/habits are not binary variables, they both seem like they would be relatively continuous (you can drink 1 can of soda a day, or 2, or 3, or 4, or 5, or anywhere in between for that matter; and you can be very concerned and conscientious about your life, or concerned, or somewhat concerned, or a little concerned), and so you would expect dose-response relationships. There's no less reason to expect linearity from soda/health than soda/telomeres.
(If it was as easy as all that to refute healthy user bias and other confounds, they wouldn't be such a big deal...)
However I would not expect it to then give a clear linear link between specific consumption and telomere length as shown in this study, as the confounding factors are unlikely to be directly linked to exactly how much you consume:
"we found that each daily 8-ounce serving of sugar-sweetened sodas was linearly associated with shorter telomeres, roughly equivalent to 1.9 additional years of aging, independent of sociodemographic characteristics and health-related variables"