`juju deploy $LOCAL_CHARM` is very slow on arm64

Back in August 2022, I reported that `juju deploy`'ing a local charm was slow:

```
$ ll -h lxd-cloud-cell_ubuntu-22.04-amd64.charm lxd-cloud-cell_0+git.d27fa0c_amd64.snap
-rw-rw-r-- 1 sdeziel sdeziel 20M Aug 2 22:02 lxd-cloud-cell_0+git.d27fa0c_amd64.snap
-rw-r--r-- 1 sdeziel sdeziel 5.8M Aug 17 19:17 lxd-cloud-cell_ubuntu-22.04-amd64.charm

$ time juju deploy ./lxd-cloud-cell_ubuntu-22.04-amd64.charm cell-b --resource snap=lxd-cloud-cell_0+git.d27fa0c_amd64.snap -n 3 --to 22,23,24
Located local charm "lxd-cloud-cell", revision 10
Deploying "cell-b" from local charm "lxd-cloud-cell", revision 10 on jammy

real 1m31.175s
user 0m3.435s
sys 0m5.517s
```

I don't know which Juju version that was but now with Juju 2.9.34, this is way worse even with a smaller charm not using any resource:

```
sdeziel@shell01:~$ ls -lh ./lxd_ubuntu-20.04-amd64.charm && time juju deploy --debug -m test ./lxd_ubuntu-20.04-amd64.charm
-rw-r--r-- 1 sdeziel sdeziel 11M Oct 5 21:00 ./lxd_ubuntu-20.04-amd64.charm

14:46:34 INFO juju.cmd supercommand.go:56 running juju [2.9.34 20b8212e62c5d122e63141825a7733bf66c4c5a9 gc go1.18.5]
14:46:34 DEBUG juju.cmd supercommand.go:57 args: []string{"/snap/juju/20511/bin/juju", "deploy", "--debug", "-m", "test", "./lxd_ubuntu-20.04-amd64.charm"}
14:46:34 INFO juju.juju api.go:78 connecting to API addresses: [172.17.40.60:17070]
14:46:34 DEBUG juju.api apiclient.go:1153 successfully dialed "wss://172.17.40.60:17070/api"
14:46:34 INFO juju.api apiclient.go:688 connection established to "wss://172.17.40.60:17070/api"
14:46:34 INFO juju.juju api.go:78 connecting to API addresses: [172.17.40.60:17070]
14:46:34 DEBUG juju.api apiclient.go:1153 successfully dialed "wss://172.17.40.60:17070/model/bab6b8e8-3e8c-40e0-8eec-118ea6a15196/api"
14:46:34 INFO juju.api apiclient.go:688 connection established to "wss://172.17.40.60:17070/model/bab6b8e8-3e8c-40e0-8eec-118ea6a15196/api"
14:46:34 DEBUG juju.core.charm computedseries.go:27 series "focal" for charm "lxd" with format 2, Kubernetes false
14:46:34 DEBUG juju.core.charm computedseries.go:27 series "focal" for charm "lxd" with format 2, Kubernetes false
14:46:34 INFO cmd charm.go:384 Preparing to deploy local charm: "lxd_ubuntu-20.04-amd64.charm"
14:53:43 INFO cmd charm.go:406 Located local charm "lxd", revision 4
14:53:47 INFO cmd charm.go:236 Deploying "lxd" from local charm "lxd", revision 4 on focal
14:53:55 DEBUG juju.api monitor.go:35 RPC connection died
14:53:55 DEBUG juju.api monitor.go:35 RPC connection died
14:53:55 INFO cmd supercommand.go:544 command finished

real 7m22.082s
user 0m0.501s
sys 0m0.414s
```

In both cases, the Juju controller is a local LXD container so the network upload of the charm shouldn't be the source of the slowness.

*Important note*: the slowness is only observed on an arm64 controller deploying to amd64 machines. The same deployment from an amd64 controller is way faster (~1 minute).