Remove infra_ip_first and infra_ip_last from UserSpecifiedRackNetworkConfig and RackNetworkConfig

[andrewjstone]

These are unused except for validation in wicket. This is overly constraining as pointed out in a comment in the type:

// TODO: #3591 Consider making infra-ip ranges implicit for uplinks
/// First ip address to be used for configuring network infrastructure

[jgallagher]

I took a stab at this and unfortunately hit a bit of "dog chases own tail" plus one legitimate use of these that I'm not sure how to untangle.

I confirmed these fields are not used by anything that reads the RackNetworkConfig from the bootstore. That's a strong indicator we should remove these fields, or at least move them somewhere else if they're still needed.

The one place we act on these is in the RSS -> Nexus handoff: they're used to create the initial address lot block with the special name INFRA_LOT ("initial-infra"):

omicron/nexus/src/app/rack.rs

Lines 305 to 322 in 72562b7

	let address_lot_name = Name::from_str(INFRA_LOT).map_err(|e| {
	Error::internal_error(&format!(
	"unable to use `initial-infra` as `Name`: {e}"
	))
	})?;
	let identity = IdentityMetadataCreateParams {
	name: address_lot_name.clone(),
	description: "initial infrastructure ip address lot".to_string(),
	};
	let kind = AddressLotKind::Infra;
	let first_address = rack_network_config.infra_ip_first;
	let last_address = rack_network_config.infra_ip_last;
	let ipv4_block =
	networking::AddressLotBlockCreate { first_address, last_address };
	let blocks = vec![ipv4_block];

The only code that reads this lot by this name is the sync_switch_configuration background task, which only reads this address lot to construct a RackNetworkConfig to write to the bootstore:

omicron/nexus/src/app/background/tasks/sync_switch_configuration.rs

Lines 1286 to 1300 in 72562b7

	let blocks = match self.datastore.address_lot_blocks_by_name(opctx, INFRA_LOT.into()).await {
	Ok(blocks) => blocks,
	Err(e) => {
	error!(log, "error while fetching address lot blocks from db"; "error" => %e);
	continue;
	},
	};
	// currently there should only be one block assigned. If there is more than one
	// block, grab the first one and emit a warning.
	if blocks.len() > 1 {
	warn!(log, "more than one block assigned to infra lot"; "blocks" => ?blocks);
	}
	let (infra_ip_first, infra_ip_last)= match blocks.get(0) {

There are several problems with this:

1. As mentioned above, the bootstore doesn't use this information. So we have a special address lot block name that's only used to construct a value that isn't consumed by anything.
2. A single address lot block is too restrictive; this is Single address lot block for infra IPs is insufficient #8179.
3. Dogfood already has another address lot block, which causes sync_switch_configuration to (needlessly, it seems!) complain that it's discarding information when constructing the RackNetworkConfig; this is sync_switch_configuration complains about multiple address lot blocks in the initial-infra address block #10182.

All that said, just because nothing consumes this special address lot by name, doesn't mean that nothing consumes address lots at all: they're used to validate new uplink addresses (although this is itself buggy, at least for addrconf uplinks: this is #10103), and some datastore methods related to switch port settings join against the address lot table for reasons I don't currently understand (possibly all related to validation of addresses?). So I don't think we can just remove these fields from the RSS -> Nexus handoff without breaking something here.

We probably need some assistance from @rcgoodfellow / @internet-diglett / others to understand what's going on here and how we can shift this information around to get it out of the low-level RackNetworkConfig / bootstore, since we know it's unused there.

[internet-diglett]

We might be interested in removing the requirement for address-lots altogether, since we don't do any sort of auto-allocation of infrastructure IP addresses. I'm not sure it's providing any unique value to the process of rack network configuration.