Dunno what the CDT answer would be (i.e., whether a given individual is more likely to die when cycling one mile or when driving one mile, all other things being equal -- I suspect it depends on where you are, anyway), but it seems obvious to me that the RDT answer (i.e., if x% of the miles travelled in a town are cycled and (100 - x)% are driven, whether the expected total number of people dying in road accidents per mile travelled would increase or decrease with x) is that cycling is safer. And if you only care about your own life and are a CDTist, you shouldn't vote unless there's a non-negligible probability of your vote deciding the election, you should defect in the PD, etc.

I don't live my life as a CDTing agent (nor does anyone else, for that matter), but of the examples you listed, none seem particularly problematic for CDT. As far as I am concerned, not voting usually is the correct decision when your own decision doesn't significantly influence others' decisions (which for me, in a national election, it doesn't). Also, actual PDs are extremely rare in everyday life, rather than, say, the indefinite iteration (with error) variety (and CDT seems to handle the latter just fine, as far as I know).