Use GitHub actions to build and publish Rust docs

[tiziano88]

This is not feasible on GCP because we don't have a way of
authenticating only post-merge workflows.

GitHub actions instead get provisioned with a read-only token for pull
requests, and with a read + write token for merge.

[conradgrobler]

LGTM

[daviddrysdale]

This seems to be generating gh-pages commits for things that don't affect the generated docs.

~/src/oak{master}:git log -n3 --oneline upstream/gh-pages
4d09e945bc42 (upstream/gh-pages) Update gh-pages from bc98624bdcf6629c4f3ef4aa139bd4a0892f3744
cf1d4380558d Update gh-pages from 6ceb81cabc00decf21e2875bbeac9d7fabd8b1fe
4e733345b21e Update gh-pages from 9ccaae8f7b94006b8f23f5c9d9ca8f88b3fe66f5
~/src/oak{master}:git diff --name-only 9ccaae8f7b94006b8f23f5c9d9ca8f88b3fe66f5..6ceb81cabc00decf21e2875bbeac9d7fabd8b1fe
Dockerfile
docs/INSTALL-OSX.md
scripts/build_server
scripts/run_server

[tiziano88]

Yes it will generate them for each commit, I imagine we could write some regexp to only trigger the workflow if any Rust file (or proto file, or BUILD file, etc...) changed, but I don't think it's worth the effort, the point of this is that it is now automated and we don't need to deal with it.

[tiziano88]

Ah but I see your point, now we have random commits like this: cf1d438 which AFAICT are just caused by some non-determinism in the generation.

[daviddrysdale]

Yeah, I didn't think cargo-doc was deterministic.

Having a commit on gh-pages for every single commit on master, whether it affects the docs or not, is pretty ugly. However, I guess in practice it's unlikely to cause problems (we don't commit often enough that the gratuitous cache invalidation is a problem, and anyone trying to (say) track down a historical API changes is more likely to use the source in master than the docs in gh-pages).

we could write some regexp to only trigger the workflow

Something along these lines was deleted in d7d8dbd