support-bundle storage: address lingering TODOs around transition from StorageManger to config reconciler

[jgallagher]

This PR makes two changes; the first is small and does affect prod (but should be very obviously correct), and the second is large but only affects tests:

1. Replace LocalStorage::dyn_datasets_config_list() -> Result<DatasetsConfig, _> with LocalStorage::dyn_dataset_config()-> Result<DatasetConfig>. The former returned an entire DatasetsConfig, which no longer exists inside OmicronSledConfig, so the prod implementation of the method had to create one. But the sole caller really only wanted that to look up a single dataset, so replace the method with one that looks up a single dataset.
2. Convert the tests from using a LocalStorage backed by StorageManagerTestHarness to being backed by sim::Storage. This has a couple pros and one con: pros are that we remove a reference to StorageManager (which is no longer used in prod, but this isn't obvious - ideally we could get rid of all the test consumers and then remove it to avoid confusing) and that the tests can now run on non-illumos systems; con is that we're no longer exercising true ZFS operations in these tests. But I think the point of the LocalStorage trait is that we don't need to exercise ZFS itself, so this con seems pretty mild to me?

Using sim::Storage required adding some tracking of a (fake) mounted property to the "datasets" it keeps track of in memory. I think this is okay? But the nested dataset / dataset split in the APIs is still confusing to me so I could easily have screwed something up in here.

[smklein]

I know this is a nitpick, but can you humor me?

Suggested change

	// in-memory property for whether this dataset is mounted; typically `true`,
	// but can be explicitly set to false for some tests
	// in-memory property for whether this dataset pretends to be mounted; typically `true`,
	// but can be explicitly set to false for some tests

Just to clarify: This boolean only changes how this gets reported; it is not used in any real logic beyond "it's a sticky bool of whatever last value you set".

[jgallagher]

Just to clarify: This boolean only changes how this gets reported; it is not used in any real logic beyond "it's a sticky bool of whatever last value you set".

Totally correct, and I wonder if we basically make this the comment instead? Something like

// In-memory flag for whether this dataset pretends to be mounted; defaults
// to true.
//
// Nothing in the simulated storage implementation acts on this value; it is merely
// a sticky bool that remembers the most recent value passed to
// `nested_dataset_set_mounted()`.