Bin packing visibility

Currently it’s possible to have plenty of virtual capacity but end up in a state where the rack still rejects new resource creation because it doesn’t have enough contiguous resources (e.g. enough CPUs on any sled) to be able to place the resource. 

We may wish to add some types of constraints to prevent some silos from exhausting a certain class or size of resource. For instance, we may provide the ability to specify that a silo may only have 1 64 vCPU instance. This is a bit of an arbitrary example, but this would hopefully ensure that for the silos that needed that class of resource they could continue to have access to it.

1. We need the UI to adequately communicate why an instance cannot be started if there is not sufficient capacity on the available sleds.
2. We need information represented in the UI about current maximum provisionable resource sizes. E.g. there is a maximum contiguous capacity of CPU to create an instance with 32 vCPUs

Related:

https://github.com/oxidecomputer/omicron/issues/5104#issuecomment-1953522778

No file, test, or entry point is named. Start by locating the instance-start and instance-creation UI, then review the related Omicron issue comment for capacity context. Done means the UI explains capacity-related start failures and shows current maximum provisionable resource sizes.