1. Is that for each candidate?
Candidates you don't like or don't know, in range voting, you would generally leave blank. The ballot would be counted as voting zero for that candidate. In a range election for a single office, probably a great deal of people would just bullet vote for one candidate, and that's fine. The extra votes would mostly come in to play in multi-candidate elections, where people who support minor parties could express their support by rating their favorite candidate but also support a majority party candidate for victory. People wouldn't be required to pick one or the other.
Probably also this would lead to more centrist candidates winning.
In my experimental system, you'd probably want to rate at least as many candidates as there are seats, just like a bloc election. As to how many seats, that's not for me to say. More seats would be more proportional, but more confusing because there would be more candidates. In general I wouldn't recommend it be used to select any more than five seats. In Ireland, for instance, they always use 3-5 seat districts for their elections.
2. Does it just apply to the general election or is it for the direct primaries also?
You could just use it in the primaries, use it to select two candidates for a runoff in the general. Sort of like the jungle primary method, but mitigating some of the spoiler effects that causes (like the situation in California, where people were fearing that two Republicans might be selected for the runoff due to splitting among Democrats). Although I'd recommend the ballots be reweighted to choose the opponent, so that ballots that voted for the winner counted less. Otherwise the majority party would likely always be able to always pick both candidates for the runoff.
For my experimental method, or another proportional method, you would likely want to use it in both primaries and the general. At least the general, otherwise of course the system wouldn't be proportional and there'd be no purpose.