ci: install from the runner's package lists instead of refreshing them first

[Paururo]

The timeout from #22 did its job on the very next run. Both jobs stopped after ten minutes instead of three and a half hours, and named the step:

##[error]The action 'Install build dependencies' has timed out after 10 minutes.

That turned a mystery into a measurement.

Where it actually hangs

Reading the failing job's log: the hang is in apt-get update, not in the install. It got through nine of roughly thirty repositories in ten minutes and never reached cmake.

Those thirty include Microsoft's azure-cli and prod feeds and Google's Chrome feed. This project has no use for any of them; they ship with the runner image and get queried on every run.

The change

The image already carries package lists, and cmake 3.22.1 has been in jammy since release. So install from those lists, and only refresh if a package cannot be found:

sudo apt-get install -y cmake ||
  { sudo apt-get update && sudo apt-get install -y cmake; }

The slow path survives for the day a list really is stale. It simply is not the path taken every time.

Applied to all four apt steps: both CI jobs, the docs social-card libraries, and the Ubuntu leg of the release matrix.

Expected side benefit

That step was the single most expensive thing in CI:

Job	apt step	Whole job
test	193s	5.2 min
GUI (Linux)	75s	3.2 min

Installing these packages takes seconds once the lists are not being rebuilt from thirty remote repositories.

Verified

The shell of each step was syntax-checked with bash -n, the fallback branch was exercised both ways, and the timeouts from #22 are untouched: 8 of 8 jobs still carry a ceiling.