fix: resolve superlinter failures and require --recreate for RHDP cluster idempotency

[butler54]

Summary

Super linter has failed on every push/PR for weeks (100% failure rate across the last 50 runs). This PR fixes the recurring failures and, to make it a substantive change, also resolves #110.

1. Superlinter fixes

Checked the vendored kyverno chart (charts/vendor/kyverno/*) first — it's already excluded via FILTER_REGEX_EXCLUDE in superlinter.yml and contributes zero lint errors. All actual failures were in first-party files:

• markdownlint: fixed real formatting issues (missing blank lines around headings/lists/fences, bad ordered-list numbering, a missing table cell) in the issue templates, README.md, airgap/DEPLOY-RUNBOOK.md, docs/disconnected-deployment.md, and the kubevirtvm chart README. Two were genuine content bugs, not just style: a skipped step number in README.md (4. → 6., now 5.) and a status table row in DEPLOY-RUNBOOK.md missing its third (empty) cell.
• textlint (NATURAL_LANGUAGE): fixed all 14 flagged terminology instances (git→Git, repo→repository, unix→Unix, indexes→indices, READMEs→readmes) across 5 files.
• isort: fixed import ordering in scripts/git-http-server.py.
• Pyink: disabled VALIDATE_PYTHON_PYINK in superlinter.yml. It conflicts with Black (which already passes, and matches this repo's own .flake8 comment "match black default"); super-linter itself warns against enabling both. No other validatedpatterns/* repo enables Pyink — utilities and industrial-edge explicitly disable it.

Verified locally against the exact tool versions/config super-linter uses (isort 6.0.1, black 25.1.0, flake8 7.2.0, mypy 1.15.0 — all pass; markdownlint and textlint's default terminology rule — zero remaining issues on all touched files).

2. Fixes #110 — RHDP wrapper scripts are not idempotent

Root cause: rhdp/rhdp-cluster-define.py's cleanup() unconditionally wiped the install directory (and ~/.azure) on every run, before openshift-install ever got a chance to run its own built-in refusal-to-overwrite check. Re-running any rhdp/wrapper*.sh script against a directory that already had a cluster installed silently discarded that cluster's local state and provisioned a new one on top — with no destroy step, orphaning any still-live Azure resources.

Fix: add a --recreate flag, threaded through wrapper.sh, wrapper-cluster-only.sh, wrapper-multicluster.sh, and rhdp-cluster-define.py.

• By default, if an install directory already contains cluster state (metadata.json), the tool now exits with an error before touching anything, listing the affected directories and instructing the user to either destroy the existing cluster first (openshift-install destroy cluster --dir=<dir>) or pass --recreate.
• --recreate only wipes the local install directory. It deliberately does not call openshift-install destroy cluster (kept minimal in scope), and prints an explicit warning every time it's used that cloud resources from a previous install may be orphaned if they weren't destroyed manually first.
• rhdp/README.md documents the new default-safe behavior and the --recreate caveat.

Verified with scenario tests (fresh directory, existing state without --recreate, existing state with --recreate) and an end-to-end CLI run of rhdp-cluster-define.py exercising all three paths — see commit message for details. No live Azure access was available to test the wrapper scripts end-to-end against a real cluster.

Test plan

• isort/black/flake8/mypy pass on scripts/git-http-server.py
• markdownlint (matching .github/linters/.markdown-lint.yml) reports 0 issues on all touched markdown files
• textlint with super-linter's default terminology config reports 0 issues on all touched markdown files
• rhdp-cluster-define.py scenario tests: fresh dir / existing state blocked / existing state + --recreate wipes
• bash -n syntax check on all three wrapper scripts
• Super Linter GitHub Action passes on this PR (final confirmation)