rfc20: drop the attributes section

[garlick]

Problem: the Rv1 attributes section is used by the early Fluxion queue implementation to store the assigned queue name in the system.scheduler key, but queues are now handled differently per RFC 33.

Drop the attributes section as discussed in #401

[grondo]

In case it wasn't clear in #401, this section is still required by Fluxion for multi-queue restart (on restart presumably Fluxion does not have access to jobspec to get the queue name, which for some reason it needs). I'll open an issue in flux-sched and refer to this PR.

[garlick]

Yes that is what I was thinking. Move RFCs forward, let code catch up. And it should have access to the jobspec...oh maybe you meant in the place where it's using the queue name.

[grondo]

FYI: flux-framework/flux-sched#1108

[grondo]

And it should have access to the jobspec...oh maybe you meant in the place where it's using the queue name.

Yeah, I'm guessing this is in the scheduler "hello" protocol. I think only R objects are returned by this protocol. Stashing the queue name in R may have been an optimization to avoid fetching the jobspec for every R? (I actually don't know, just guessing)

[garlick]

Since flux-framework/flux-sched#1108 was resolved, I think fluxion no longer requires the attributes section of Rv1. (As a test I installed the latest fluxion on my test system with queues enabled and verified that no attributes section appears in the R assigned to a test job). So if there was a concern about fluxion's design requiring that section, it appears to have been addressed :-)