riscv64 support almost tripled snap build times

[Martin Kalcok]

Comparing run times of `build` step in the CI before [0] and after [1] merging of support for rscv46 builds [2], shows that it almost tripled.

I assume that it's because all the python dependencies for docs are built from source now. I'm not sure if there's a better way to do this, but we should at least look into it, as the build time change is very noticeable.

[0] https://github.com/canonical/microovn/actions/runs/9285575573/job/25550818064
[1] https://github.com/canonical/microovn/actions/runs/9321247482/job/25662447092
[2] https://github.com/canonical/microovn/pull/134

[Frode Nordahl (fnordahl)]

The motivation for the current approach is to ensure we detect missing build dependencies in our gate, avoiding post-merge build failures further down in our pipeline.

There is an simultaneous effort ongoing to add a gate check in the upstream documentation repository [3]. Once we have that we can opt to relax this again in our repository.

Another issue is that there is currently no pinning nor upper constraints in place for the Python dependencies used to build the documentation, so in my mind we are sitting on a ticking bomb waiting for some direct or indirect dependency to introduce breaking changes.

There is no simple and clear path to address that, but our build pipeline only needs to render the doc source, so perhaps we could do with a subset of the dependencies and even source them from Ubuntu packages, which would give us a branch lifetime worth of vendored and security maintained dependencies.

3: https://github.com/canonical/sphinx-docs-starter-pack/issues/238