Roughly 3 changes:
1. Skip initializing any zones we already know are running (both from
sled-agent's perspective and as reflected via `zoneadm`)
2. Log any errors `service_put` would return on the server-side.
3. Bump file-based zpools to 15G

Lack of space was causing service initialization to fail on
single-machine deployments using the `create_virtual_hardware` script.
(3) should fix this.

Because of `ENOSPC` errors, we were running into some future
cancellation issues which #3707 addressed. But actually determining the
underlying cause was a bit difficult as `RSS` was timing out (*) on the
`services_put` call and so never got back the error. (2) should
hopefully make this easier to catch in the future, e.g.:
```
04:12:31.312Z ERRO SledAgent: failed to init services: Failed to install zone 'oxz_clickhouse_32ce0d6f-38fa-41e7-b699-f0af4f4f1127' from '/opt/oxide/clickhouse.tar.gz': Failed to execute zoneadm command '
Install' for zone 'oxz_clickhouse_32ce0d6f-38fa-41e7-b699-f0af4f4f1127': Failed to parse command output: exit code 1
    stdout:
    A ZFS file system has been created for this zone.
    INFO: omicron: installing zone oxz_clickhouse_32ce0d6f-38fa-41e7-b699-f0af4f4f1127 @ "/pool/ext/24b4dc87-ab46-49fb-a4b4-d361ae214c03/crypt/zone/oxz_clickhouse_32ce0d6f-38fa-41e7-b699-f0af4f4f1127"...
    INFO: omicron: replicating /usr tree...
    INFO: omicron: replicating /lib tree...
    INFO: omicron: replicating /sbin tree...
    INFO: omicron: pruning SMF manifests...
    INFO: omicron: pruning global-only files...
    INFO: omicron: unpacking baseline archive...
    INFO: omicron: unpacking image "/opt/oxide/clickhouse.tar.gz"...
    stderr:
    Error: No space left on device (os error 28)
    sled_id = 8fd66238-6bb3-4f08-89bf-f88fc2320d83

```

(*) speaking of which, do we want to increase that timeout from just
60s? it's not like any subsequent requests will work considering they'll
be blocked on a lock from the initial request. and in failure cases like
this the subsequent requests will timeout as well, leaving a bunch of
tasks in sled-agent all waiting on the same lock to try initializing.
might be worth switching to the single processing task model rack/sled
initialization use.

Finally, (1) is to address the case where we try to initialize a zone
that's already running. That's sometimes fine as we'll eventually
realize the zone already exists. But in the case of a zone with an OPTE
port, we'll run into errors like so:
```
    error_message_internal = Failed to create OPTE port for service nexus: Failure interacting with the OPTE ioctl(2) interface: command CreateXde failed: MacExists { port: "opte12", vni: Vni { inner: 100 }, mac: MacAddr { inner: A8:40:25:FF:EC:09 } }
```
This happens because the
[check](https://github.com/oxidecomputer/omicron/blob/ebd3db27f6bbd08af3194cae84b0efc6b6e7248d/illumos-utils/src/zone.rs#L282C27-L282C27)
for "is zone running" comes after we do all the work necessary to create
the zone (e.g. creating an opte port). (1) updates
`initialize_services_locked` to cross-reference sled-agent and the
system's view of what zones are running so that we don't try to do the
unnecessary work.

---------

Co-authored-by: Sean Klein <sean@oxide.computer>