e2e: move per-test resource cleanup into bats teardown() #2583

Problem

E2E tests under hack/e2e-apps/*.bats perform resource cleanup inline at the
end of each @test body (e.g. kubectl delete vminstance ... after the wait
assertions). When an earlier kubectl wait or until step fails, the test
exits before reaching the delete lines, so resources leak into the cluster
and can pollute subsequent retries or unrelated app tests sharing the
tenant-test namespace.

Surfaced in #2555 (comment)
on hack/e2e-apps/vminstance.bats, but the same pattern exists in most
e2e-apps/*.bats files.

Proposal

Move resource deletion into a framework-level teardown() (bats hook that
always runs, even on test failure). Track created resource names in a
shell-array or temp file in setup()/test body, and have teardown()
iterate them with kubectl delete --ignore-not-found --timeout=....

Keep test bodies focused on create + assert; cleanup becomes guaranteed.

Scope

• Audit all files under hack/e2e-apps/ for inline kubectl delete at
  end-of-test.
• Introduce a shared cleanup helper (likely in hack/e2e-apps/ as a sourced
  .sh) that test files can register resources with.
• Migrate each app's .bats file individually so churn is reviewable.

Out of scope

• Changing the actual create/assert logic of any test.
• Reworking cozytest.sh runner semantics.

Acceptance criteria

• A failing kubectl wait in any e2e-apps/*.bats test still removes the
  resources it created before the next test runs.
• No inline post-assertion kubectl delete calls remain in test bodies.