fix: wait for operator readiness before creating Instrumentation

[karthikeyangs9]

Problem

Customers on slower clusters saw the setup script fail with:

[INFO] Attempt 1 of 5 to create instrumentation...
[WARN] Attempt 1 failed. Waiting before retry...
... (x5) ...
[ERROR] Failed to create instrumentation after 5 attempts.

The Instrumentation resource is gated by the OpenTelemetry Operator's mutating admission webhook, which only serves once the operator pod is Ready. The script gated readiness on a blind sleep 30, then ran kubectl apply -f instrumentation.yaml 2>/dev/null. On Autopilot the operator takes longer than 30s to come up (it provisions a node first), so every apply attempt raced an unready webhook — and 2>/dev/null discarded the real error, so the failure was undiagnosable.

Changes

• install_operator: replaced the blind sleep 30 with kubectl rollout status deployment/opentelemetry-operator --timeout=${OPERATOR_READY_TIMEOUT:-180s}. Returns the moment the operator is Ready (faster than a flat 30s on healthy clusters) and blocks up to the timeout on slow ones.
• create_instrumentation: waits on the operator deployment before applying, and captures stderr (2>&1) so a persistent failure surfaces the real API-server error (e.g. failed calling webhook ... connection refused) instead of a blind "Attempt failed". Also moved the "try manually" hint before log_error (which exit 1s) so it actually prints.

OPERATOR_READY_TIMEOUT is overridable for very slow clusters.

Tests

• New tests/test_create_instrumentation.py (3 tests): waits before apply, surfaces the real error on persistent failure, succeeds on happy path.
• tests/test_install_operator.py: added a readiness-wait assertion.
• Verified end-to-end on a kind cluster: with both fixes, instrumentation now succeeds on the first attempt (previously it failed attempt 1 and scraped by on attempt 2).

Note for the diagnosable-but-unfixable case

If the operator never becomes Ready (ImagePullBackOff, Pending, CrashLoopBackOff, x509), no wait can fix that — but these changes now surface the real reason instead of failing silently. Check kubectl describe pods -n <ns> -l app.kubernetes.io/name=opentelemetry-operator.