blueprint-execution: Can we remove the "list sleds from db" step?

[jgallagher]

Prior to the execution steps that send requests to sled-agents, blueprint execution currently queries the sled table for all in-service sleds:

omicron/nexus/reconfigurator/execution/src/lib.rs

Lines 281 to 283 in 31767f6

	let sleds_by_id: BTreeMap<SledUuid, _> = datastore
	.sled_list_all_batched(&opctx, SledFilter::InService)
	.await

This seems a little confusing - doesn't the blueprint being executed already list the sleds that it believes to be in service?

In practice, the returned list of sleds is used for two things:

1. In the three PUT-to-sled-agent steps, we use sled.sled_agent_address() (this comes from the ip+port columns in the sled table)
2. In the deploy_dns step, we use sled.is_scrimlet() to find the scrimlets to set up DNS entries for services in the switch zones

I think we could easily address 1 - we could either infer a sled's underlay address from the underlay IPs of any zones it contains plus the known port sled-agent listens on, or we could add the sled-agent underlay addresses to the blueprint explicitly.

I'm a little less sure about 2 - should the planner be aware of which sleds were scrimlets at the time of planning? Will it need to do this as we get into live update?