staging: Deploy latest nixguix lister and new vcs directory loaders

Contains:

• New hg, git, svn directory loader able to ingest those vcs trees
• New swh.loader.{core,git,svn,hg} version
• New swh.lister version

Plan:

• Tag modules and build packages
  • v5.6.1: lister
  • v5.6.0: loader.core
  • v2.4.0: loader.git
  • v3.3.0: loader.hg
  • v1.8.0: loader.svn [1]
  • v5.6.2: lister
  • v5.6.3: lister
  • v5.7.1: loader.core
• Build images
  • swh-lister
  • swh-loader-savecodenow
  • swh-loader-git
  • swh-loader-mercurial
  • swh-loader-svn
  • swh-lister
  • swh-lister
  • swh-loader-savecodenow

staging:

• swh-charts: Add new Directory loader deployment (consuming the new queues [1])
• swh-charts: Update image versions to deploy
• scheduler0.staging 'scheduler' db: Clean up listed origins with visit_type 'directory'
• Deploy
• scheduler0.staging: Upgrade impacted python3 packages
• scheduler0.staging: Register new task types [2]
• scheduler0.staging: Restart swh-scheduler-schedule-recurrent service (to pick up the new task type)
• Checks
  • listing ok
  • ingestion ok? so far, NO!
    • 1. no, wrong docker image used in the chart (fixed)
    • 2. no, visit type and task name discrepancy makes for no scheduling (fixed through sql query in db)
    • 3: no, mercurial queue name is not ok (fixed)
    • 4. no, the lister does not propagate the reference it read from the upstream manifest (fixed)
    • 5. no, the tarball directory loader's visit type is inconsistent with the task name, scheduling issue again (fixed)
  • scheduled origins are ingested (some work, some don't, sentry does catch those)
    • svn-export
    • hg-checkout
    • git-checkout
    • tarball-directory
  • ingestion on those origins is ok

[1] swh.loader.{git.tasks.LoadGit,svn.tasks.LoadSvn,hg.tasks.LoadHg}Directory

[2] load-{git,svn,hg}-directory

[3] debian build is unhappy though, not that it's too much blocking for staging

Refs. swh/meta#4979 (closed)