[nexus] Add DataStore::instance_fetch_all

[hawkw]

In order to implement the instance-update saga ongoing in #5749, it's necessary to have a way to query the database for the state of an instance, its active VMM, its target VMM, and active migration in a single, atomic query. This will be used as a snapshot of the instance state that the update saga will operate on.

This commit adds an implementation of such a query in the form of a DataStore::instance_fetch_all method, returning an InstanceSnapshot struct that bundles together all these records. We do this by LEFT JOINing the instance table with the vmm and migration table on the instance's various ID columns, with a query that should look something like:

SELECT * FROM instance
LEFT JOIN vmm
    ON ((instance.active_propolis_id = vmm.id)
    OR (instance.target_propolis_id = vmm.id))
LEFT JOIN migration ON (instance.migration_id = migration.id)
WHERE instance.id = instance_id

Ideally, we would instead do two subqueries that join instance with vmm on the active and destination IDs separately and return a single row which contains both in a way that can be distinguished in the result set, but I wasn't able to figure out how to do that in Diesel. Instead, the query may return two rows, if the instance has both an active and target VMM, and iterate over the result set to find both VMMs in Rust code. This is a bit uglier, but it works fine.

[gjcolombo]

Yeah, this is probably the best way of doing this while coloring inside the lines Diesel has set out. (I could imagine having some kind of CTE that conjures up a temporary table that contains the VMM IDs and a "sequence" column that can be used to order the returned rows unambiguously, but I've found it very hard to do this kind of thing without dropping out to raw SQL.)

[hawkw]

That's sort of what I thought. Writing the CTE seemed like it would be enough of a pain that I didn't really think it would be worth the effort, but we can revisit this later.