To me the whole question of how much to cite is an artifact of trying to use lineage-linked software as a tool for recording what we find in evidence. It took me years to figure out that these programs are designed to keep track of the material which we have already 'proven' or at least have concluded belongs to the same person. This may sound obvious, but I was trained to handle data very differently, so to me the whole data-entry process for a lineage-linked program is backwards from the way it should be.
In a source-based process, it works like this. You keep a research journal of what you collect, when you found it and where you found it. Let's say you have an oral history, because that's the kind of family history material which is most like the data I was trained to collect. You interview Aunt Jane. You record what she said, and then you go home and rewrite your notes and/or listen to your sound recording or video, and you make a transcript. Then you break that transcript down into all the bits of evidence that Aunt Jane told you, which some people call facts. But in reality, it's all stuff that you only know because "Aunt Jane says so", and to remind ourselves of this, some family historians prefer to call them "assertions".
From this perspective, the answer to "where do you put the citations" is automatically "everywhere". You've collected this one source, and you've started to extract the assertions in it. You haven't done any analysis of it, or proven anything about which person it belongs to. You might have an idea of where it belongs because that's why you collected it in the first place, but that idea isn't proof. It may make up part of your proof later, but you aren't there yet. So you have to create a citation -- because you have to record the source -- for all the individual assertions now, as you are extracting them -- otherwise, how do you keep track of their source while you are doing the analysis? (In the old days, we did this on paper, and we had boxes of index slips -- and we indexed everything. Every bit, each on its own slip, with each slip having a reference to where it came from, in case we discovered the source was crap and we had to throw it out. Each slip would contain an assertion in context, which I think is similar to what others call "persona" records. Slips that we thought belonged together got filed together, but nothing was attached in the same way evidence is attached to a person in a lineage-linked database. It was all much more fluid.)
This is why of all the lineage-linked software that I looked at, I chose Family Historian. It has auto-source citation. It was the only program that I found that would let me extract assertions from a source and post the data to the person I thought it belonged with, that would add the citations automatically as I went along. This is still not ideal, because it still encourages me to make up my mind about what person these assertions belong to in advance of actually doing the work to prove my conclusion, but it is better than working from the person-centric mindspace, where you add the data to the person first, and then try to figure out "where to put the citations" -- which results in what one user here called a "pedigree with source citations hanging off it". (See Mat's question What tools exist for collecting and managing evidence?)
I can't give you a good answer which is specific to Gramps, because when I looked at it, I was trying to figure out how I can make it work as a source-centric model of data handling, and I couldn't do it. So I haven't looked at it for very long. We have a lot of wonderful tools at our disposal, but IMHO they aren't always designed to do the things we need them to do, and if you are a newcomer and don't realize what the hidden assumptions are, it is easy to shoot yourself in the foot with them.
About your census example, Cole Valley Girl says:
If I have other sources for the same assertion (e.g. birth date), I
might rank it as lower in precedence but I'd still record it as
supporting the assertion. Same if it contradicts other evidence -- the
discrepancy tells me something (even if I may not know what!).
I think it's important to record the data somewhere, but the question is, where? In a formal citation? In a local note? In your research journal?
One criterion for deciding which way to go might be the nature of the evidence itself. That is, for a death certificate, record the evidence and produce a citation for the stuff that that the document is about -- the death. But for the evidence which is not contemporary to that event, like the birth, one might put those assertions in source notes instead, along with the analysis of whether or not the birth date and place are consistent with your other evidence. But whether you create a formal source citation or not, I think it's important to keep some kind of record of what was asserted where.
In the system I was trained in, we used facing pages in the research journal. Evidence was recorded on the right-had page, and our commentary and analysis was on the left page, so it was always evident what came from the source and what was our own analysis. All of the researcher's cross-references, observations, notes for follow up, etc. all go on the facing page on the left; the informant's data goes on the right. All pages in the journal were numbered (so you can tell if pages go missing).
When we analyzed our source material, we marked up our research slips to assign the data to the appropriate categories, and filed the slips accordingly. (The program I've seen so far in the genealogy / family history/ microhistory realm that seems the friendliest to this process is The STEMMA Project by @ACProctor.) Any slip would be coded so if you discovered data that was in error, you could pull it back out. (In a group process where data was shared, both the informant's name and the researcher's name would be linked to, so it would always be clear whose work was whose.)
In our paper-based system, the permanent record of our research is the research journal, in which each source collected is dated. The boxes of paper slips extracted from the information in that journal make up our working index. The lineage-linked software, with its source citations, is the machine equivalent of the index. When I started out, I was baffled to find many programs for the lineage-linked part, but no good software to replicate the crucial core of the process, the research journal, which provides the equivalent of an audit trail in accounting.
I think it really helps to write out your thoughts and mock up the whole process on paper. If you can't describe the process on paper and have it make sense, it won't be any better once you transfer the whole process to the computer. Writing it by hand is a different tactile experience than typing; it engages your brain in a different way and can give a fresh perspective.