Now it`s not possible to implement "manual cleanup" jobs which can be executed even if previous job failed

Summary

Looks like now it`s not possible to implement "manual cleanup" jobs which can be executed even if previous job failed.

Steps to reproduce

Create pipeline with manual job somewhere in the middle, e.g. build -> test -> manual cleanup -> deploy Execute pipeline in a way that test job will fail.

stages:
  - build
  - test
  - cleanup
  - deploy

build:
  stage: build
  script:
    - echo "test"

test:
  stage: test
  script:
    - echo "test"
    - exit 1

manual cleanup:
  stage: cleanup
  when: manual
  allow_failure: true
  script:
  - echo "cleanup"

deploy:
  stage: deploy
  script:
  - echo "deploy"

Example Project

https://gitlab.com/SCORE1387/sandbox

What is the current bug behavior?

It`s not possible to execute manual job when previous job failed. Example pipeline run: https://gitlab.com/SCORE1387/sandbox/pipelines/30622687

What is the expected correct behavior?

I expect that it`s possible to implement such behavior when I can run manual job even with previous job failed.