fix: dev retry interval, simulate label idempotency, roadmap, simulate concept doc

[iAlexeze]

Summary

1. Dev/prod CRD reactivation interval

retryMissingCRDs runs on a fixed ticker to detect and activate CRDs that were missing at startup. The interval was 90s unconditionally.

When a developer runs ork run locally — after orkestra applies the CRDs — there is a startup race where the CRD may not be fully registered yet. Previously the operator would sit in pending for up to 90s before the next check. Not a good first-run experience.

Fix: select the interval once at startup using utils.IsRunningInCluster():

• Not in a pod (ork run locally): 10s
• Running in a pod (production): 90s

2. Simulate — label patch no longer shows after cycle 1

FakeKubeclient.PatchLabels was recording the op on every call unconditionally. The real PatchLabels builds a label diff and returns early when nothing changed. After ownership labeling was added to generic.go, ~ labels/... started appearing in simulate output on every cycle including steady state — visually noisy and misleading.

Fix: check stringMapsEqual(base, desired) before recording. Labels show only on cycle 1 (first application), then disappear.

3. Roadmap

• Added ork lint — semantic/policy gate distinct from ork validate (schema correctness)
• Added namespaced katalogs — one Orkestra instance, teams scoped by namespace, CC shows separate panels
• Added declarative canary rollouts to longer horizon
• Clarified multi-cluster federation entry: cross-cluster composition works today via cross: over HTTP

4. ork simulate concept page

New page at documentation/concepts/simulate/index.md. The CLI reference explains how to use it; the concept page explains why it is architecturally different — same GenericReconciler that runs in production, wired to an in-memory store. Covers what it catches, what it cannot catch, the write→simulate→cluster workflow, and a comparison table with ork e2e.

---

Files changed

File	Change
pkg/kordinator/constants.go	postStartRetryInterval = 90s, postStartRetryIntervalDev = 10s
pkg/kordinator/post_start_hooks.go	Interval selected via utils.IsRunningInCluster()
pkg/kordinator/docs/04-self-healing.md	Documents dual interval
pkg/simulate/kubeclient.go	PatchLabels idempotency guard + stringMapsEqual
documentation/roadmap.md	Roadmap additions and corrections
documentation/concepts/simulate/index.md	New concept page
documentation/reference/cli/08-e2e.md	Minor reference fix

Test plan

• make ork — clean build
• ork run locally with a missing CRD — recheck fires within 10s
• In-cluster deployment — behavior unchanged (90s interval)
• ork simulate on any example — labels/ does not appear after cycle 1
• Roadmap renders correctly in docs site