feat: drop HELM_IS_INSTALL env-var dependency

[arthlr]

Summary

• Removes the HELM_IS_INSTALL env-var read in src/chartreuse/chartreuse_upgrade.py:88,105.
• Simplifies if UPGRADE_BEFORE_DEPLOYMENT and not HELM_IS_INSTALL: return → if UPGRADE_BEFORE_DEPLOYMENT: return — always behave as upgrade.
• Drops the env var from 8 test fixtures + docs.

Why

The env var was a Helm-specific install-vs-upgrade signal that doesn't exist under ArgoCD (helm template always reports IsInstall=true). The branch it gated — "in upgrade-before-deployment mode + first install + we ran migrations + ENABLE_STOP_PODS, then start pods manually" — has an unreachable subcase under ArgoCD, and the orchestrator (ArgoCD or Helm) handles pod rollout via Deployment manifests in all other paths. So always returning without start_pods() when UPGRADE_BEFORE_DEPLOYMENT=true is the correct behavior under both Helm and ArgoCD flows.

Test plan

• uv run pytest → 88 passed (same as baseline).
• uv run ruff check + ruff format --check → clean.
• grep -r HELM_IS_INSTALL src/ docs/ → only the deliberate explanatory comment remains.
• Reviewer to confirm no internal consumer relies on the env var.

Coordinated with

• Chartreuse Helm chart 6.3.0: feat(chartreuse): emit ArgoCD Sync hook when deploymentMethod=argocd (6.3.0) wiremind-helm-charts#859 — emits HELM_IS_INSTALL: \"\" always under ArgoCD mode, so this drop can land independently of that PR's merge order.

🤖 Generated with Claude Code

[arthlr]

Closing — not needed for the ArgoCD migration after all.

The chartreuse Helm chart change in wiremind-helm-charts#859 emits HELM_IS_INSTALL: \"\" (empty string) into the ConfigMap when its new ArgoCD mode is enabled. The unmodified Python tool reads that with os.environ.get(\"HELM_IS_INSTALL\", \"false\").lower() not in (\"\", \"false\", \"0\") which treats empty as `False`, so the existing branch `if UPGRADE_BEFORE_DEPLOYMENT and not HELM_IS_INSTALL: return` already fires correctly under ArgoCD. The Python tool change here was strictly cosmetic — removing dead-looking code.

Worse, dropping the conditional as this PR did introduces a regression for a documented Helm-flow configuration: with `upgradeBeforeDeployment: true` on a pure `helm install` (no prior release), `stop_pods()` scales the freshly-applied Deployments to 0, and then skipping `start_pods()` leaves them stuck at zero replicas because Helm `install` doesn't re-apply the manifests after the install completes. Wiremind has no consumers in that configuration but this repo is public and we can't see who does.

If we ever want to clean up the dead code, the safer refactor is to drop the entire `UPGRADE_BEFORE_DEPLOYMENT`-conditional skip and always call `start_pods()` — the orchestrator's later re-apply (under `helm upgrade` or ArgoCD wave-0 deployments) becomes a no-op rather than an "expected restore". Not in scope for the current ArgoCD migration; deferred.