While technical papers are generally considered to be "non-archival" and journal papers "archival", I couldn't find a clear statement about papers that appear in conference proceedings. Does it matter in which for the proceedings are published, e.g. only online, on CD/USB drive, or printed with/without ISBN?
Unfortunately, the reason you can't find a clear statement on this issue is because there isn't one. The answer depends heavily on the field and the particular conference, and there isn't necessarily an absolute answer even in a specific case. It may depend on why you care:
You may care yourself, because you want your paper to be carefully archived and made available in perpetuity. In that case, it depends on whether a long-lasting, trustworthy organization has committed to providing access forever (and whether they have archiving contingency plans in case they go out of business).
You may need to make this distinction to assist in evaluating your CV. In that case, the evaluators probably care primarily whether the papers are carefully refereed and in their final form, but the only way to know for sure is to ask them what their criteria are.
You may be wondering whether you can publish an updated version of your conference paper in a journal. This depends on the customs in your field, and it requires agreement from both the conference and the journal. The publishing agreement for the conference should specify what sorts of further publication are allowable (if it does not, then you should be sure to ask, since saying nothing suggests they do not expect any further publication at all), and the submission instructions for the journal may also address this issue. If your submission is a revision of a conference paper, you should always specify this fact clearly and give a precise reference somewhere on the first page and in the cover letter, to make sure nobody could accuse you of hiding the previous publication.
The answers to these questions can vary. For example, major theoretical CS conferences are archival in sense 1 (permanent archiving and availability) but not 2 or 3 (papers are not necessarily intended to be in their final form and definitive versions may be published elsewhere).
The editor of IEEE TRANSACTIONS ON SOFTWARE ENGINEERING asked the same question in editorial to VOL. 37, NO. 2, MARCH/APRIL 2011. He did not provide an answer, but it might be interesting to consult your university library and check the issues right after this one, I am sure there was a discussion in the journal. The archive does not seem to hold other than regular paper contributions.
My personal take on this (informed by the research field culture and research community I live in) is that a publication is archival whenever it comes with an assigned ISBN/ISSN. That makes the publication, be it a book, journal, or proceedings uniquely identifiable and thus it makes its source identifiable too. Whether that will make the publication physically, or digitally available on request is another story (publisher going bankrupt, etc.). I would consider this the minimal requirement.
Now to your concrete question regarding archival nature of conference proceedings, note, many conference proceedings come with an ISBN, or ISSN number. Even many technical report series at some universities do. In my field (CompSci/AI et al.) all the relevant top-tier conference proceedings have an ISBN. The lower tier conferences and more prominent workshop series tend to publish their proceedings and post-proceedings as Springer LNCS/LNAI series and thus get an ISBN too. I personally treat all such publication as archival and of course refrain from resubmitting such results elsewhere. This is however not the case for really small workshops (usually one-of), which only print their proceedings and bind them. In all conferences/journals I care for, these are treated as mere non-archival technical reports and thus can be further disseminated, or resubmitted to a better venue, such as a good conference. But again, your mileage might vary. In different fields, there are different habits regarding such results dissemination strategy.
One more thing to consider - are you requested to sign copyright transfer form (or sth similar, like consent to publish). If you are - then the publication is most likely archival and you can read on the copyright form if and under which conditions you can reuse the material. E.g., in the above mentioned LNCS-LNAI copyright form it is explicitly mentioned that you can re-publish it only with apprx 30% of new material (more or less standard condition in Computer Science)