Failed analysis stage does not properly update github CI status

[shonfeder]

When the analysis stage of the CI pipeline fails, it should fail the entire CI check, since no other jobs can proceed. But currently this is not the case, instead jobs with a failed analysis just linger as pending.

I believe this was one contributing factor to ocaml/opam-repository#26956

A current example is https://opam.ci.ocaml.org/github/ocaml/opam-repository/commit/09b8faba8678c783d4d7c8b89744d1a4f2693448/variant/(analysis) failing with

2024-11-28 20:11.16: New job: Analyse
2024-11-28 20:11.16: Waiting for resource in pool analyse
2024-11-28 20:11.34: Got resource from pool analyse
2024-11-28 20:11.34: Checking out commit 09b8faba. To reproduce:
                       git clone --recursive "https://github.com/ocaml/opam-repository.git" && cd "opam-repository" && git fetch origin "refs/pull/26992/head" && git reset --hard 09b8faba
2024-11-28 20:11.34: Exec: "cp" "-a" "--" "/var/lib/ocurrent/var/git/opam-repository.git-a0bc4b41ddb868605dc1500002c27cfa79faf4389fe53b6754817ea16b48d458/.git" 
                           "/tmp/git-checkout17b8f72b"
2024-11-28 20:11.37: Exec: "git" "-C" "/tmp/git-checkout17b8f72b" "submodule" 
                           "deinit" "--force" "--all"
2024-11-28 20:11.38: Exec: "git" "-C" "/tmp/git-checkout17b8f72b" "reset" 
                           "--hard" "-q" "09b8faba8678c783d4d7c8b89744d1a4f2693448"
2024-11-28 20:11.47: Exec: "git" "-c" "protocol.file.allow=always" "-C" 
                           "/tmp/git-checkout17b8f72b" "submodule" "update" 
                           "--recursive" "--init" "--no-fetch"
2024-11-28 20:11.47: Exec: "git" "merge" "-q" "--" "6dde9142a8737db80ad71a442b685538e76b185f"
2024-11-28 20:11.50: Exec: "git" "diff" "--name-only" "6dde9142a8737db80ad71a442b685538e76b185f"
2024-11-28 20:12.15: Job failed: Unexpected path "governance/README.md" in output (expecting 'packages/name/pkg/...')

But for which the CI tile appears as

[punchagan]

It looks like the analysis failing causes the latch to be stuck in a pending/ready state.

Some context on why the latch was added is in this comment and #202 #204 and #206