Refactor module loading

We've observed many issues over the years with ubuntu's apt servers,
including lots of temporary 503's.
This commit will
1) Use an in-gcp mirror for debootstrap, which should be fast (it
   doesn't run on ubuntu's infrastructure at all)
2) Add retries for all apt-get invocations during the build of both the
   os-image, and the iaas-specific builds - details below

Acquire::Retries=10 If a download fails (including transient HTTP errors
like 503), apt retries that fetch up to 10 times before giving up on
that URL. Without this, one bad response often aborts the whole apt-get
run.

Acquire::Retries::Delay=true Between retries, apt waits with increasing
delay (backoff) instead of hammering the server immediately. On Jammy
this is the “delay between retries” behavior (apt ≥ 2.0). Cumulatively
with Retries=10, you get many attempts spread over time, which helps
when snapshot.ubuntu.com is briefly overloaded rather than hard-down.

Acquire::http::Timeout=120 and Acquire::https::Timeout=120 Each
individual HTTP/HTTPS connection apt opens to a mirror can block for at
most 120 seconds before apt treats it as stuck and fails that attempt
(which can then trigger a retry if retries remain).

ai-assisted=yes
[TNZ-88995]

Add apt retries, optional mirror for debootstrap

🐛 Fix ytt comment

Turns out, the internal GCE mirror is designed only for actual vms
running in GCP, and either didn't play well with the Concourse
container's egress or doesn't server jammy.

Fixes

debootstrap --arch=amd64 jammy /mnt/stemcells/null/null/ubuntu/work/work/chroot http://gce.archive.ubuntu.com/ubuntu
I: Retrieving InRelease
I: Retrieving Release
E: Failed getting release file http://gce.archive.ubuntu.com/ubuntu/dists/jammy/Release

…irror

Use kernel.org mirror for debootstrap