k8s-e2e workflow: pin the CI cluster name instead of the derived cds-head #1040

Follow-up from the review of #1027 (loose end of #76).

In CI, actions/checkout leaves the runner at a detached HEAD, so
scripts/k8s/k3d-env.sh derives CDS_BRANCH=HEAD and the k3d cluster is
named cds-head (visible in the make k3d-down step and in diagnostics).
Functionally harmless on GitHub's single-use runners — every script sources
the same env file, so the name is consistent within a run — but it makes CI
logs and artifacts harder to attribute than e.g. cds-k8s-e2e-ci.

Suggested fix: set an explicit cluster name in the job env of
.github/workflows/k8s-e2e.yml, e.g.:

env:
  CDS_CLUSTER: cds-k8s-e2e-ci

and extend tests/test_k8s_e2e_workflow.py to pin it. Effort: small.