[type:refactor] enable external ingress adapters for Motan asset handoff - #7498

Summary

This PR introduces a clean extension point — the IngressPluginDefinition interface — so that non-core plugins like Motan can be wired in as external ingress adapters instead of being hardcoded in the kubernetes controller. It also removes the now-orphaned motan LICENSE entries from the bootstrap distribution. The change is a refactor/extension, not a behavior change for any plugin that ships in the main repo today.

What I verified and liked

• Backward compatible. IngressParser keeps a no-arg constructor that delegates to an empty list, and the k8s starter now injects ObjectProvider<IngressPluginDefinition>.orderedStream() (sorted, so @Order resolves overlap — matches the interface javadoc). With zero external definitions (the current main-repo state) the list is empty and behavior is unchanged — confirmed by the all-green k8s integration suite (k8s-ingress-http/-websocket/-dubbo/-grpc all pass, plus pr_build/rat_check/Analyze).
• Consistent dispatch priority. parse, getPluginName, and enablePluginsBasedOnAnnotations all resolve in the same order: external definition first (findFirst), then the built-in dubbo/ws/grpc/sofa chain, then the DIVIDE fallback. This means the selector created by parse (which carries the plugin name) and the plugin name used by processSelectorData / updateUpstreamFromEndpoints never disagree.
• Symmetric add/delete. The deletion path threads the external plugin name + metadata paths through the new deleteSelectorByIngressName overload, and IngressSelectorCache.remove uses the same plugin name as the selector creation. The cache put/remove keys match.
• LICENSE cleanup is verified-safe, not a compliance regression. I checked master: shenyu-plugin-motan no longer exists under shenyu-plugin, and neither shenyu-bootstrap/pom.xml nor shenyu-dist/shenyu-bootstrap-dist/pom.xml references motan. Since motan has moved to the shenyu-plugin-store repo and is no longer packaged into the bootstrap, removing the four motan entries + LICENSE-motan.txt is the correct cleanup.
• Tests are non-vacuous. testDispatchToExternalDefinitionWhenMatched sets both plugin-fake-enabled=true and PLUGIN_DUBBO_ENABLED=true and asserts only fake is dispatched — this proves the external definition wins over the built-in dubbo branch (would fail on master, which has no such concept). testExternalPluginReconcileAndDeleteUseExternalPluginName exercises the full reconcile → delete lifecycle under the external plugin name (saveOrUpdatePluginData("fake") captured, selector created/removed under "fake"). Both genuinely depend on the new code.

Non-blocking observations / suggestions

1. (Pre-existing, not introduced here) getPluginName and enablePluginsBasedOnAnnotations have no BRPC branch, while parse does. An ingress that enables only brpc would be parsed into a brpc config, but getPluginName falls through to DIVIDE and mis-routes the selector. This is a latent inconsistency in the existing code; please track it as a follow-up issue rather than folding it into this refactor.
2. Static cache concurrency (pre-existing). IngressSelectorCache.get returns the live internal List from the ConcurrentMap (not a copy), and that cache is shared by the two reconcile worker threads (IngressControllerConfiguration wires each controller with withWorkerCount(2)). Concurrent put/remove/iterate on the same ingress key can throw ConcurrentModificationException. This predates the PR, but the new metadataPaths loop slightly widens the iteration surface. Consider returning a defensive copy, or documenting the contract.
3. Design clarity. When an ingress matches an external definition, enablePluginsBasedOnAnnotations entirely skips the built-in plugin enablement (dubbo/ws/grpc/sofa). That is the intended "takeover" semantics, but it's worth stating explicitly in the IngressPluginDefinition javadoc / PR description so that a user who sets both an external annotation and a built-in one understands the built-in side is silently ignored.
4. Upstream refresh scope. updateUpstreamFromEndpoints only refreshes DIVIDE/WEB_SOCKET upstreams and now also skips external plugins. That is correct for registry-based plugins (Motan/dubbo/sofa manage their own upstreams), but if a future external definition wants k8s-endpoints-based upstreams it will need to override the path — a note on the interface would help.
5. Test readability. assertEquals(new Result(false), result) in the new test reads like "reconcile failed". Result(false) is the kubernetes-client framework's "do not requeue" signal (the method always returns it), not a failure. A one-line comment would prevent future misreads.

Conclusion

No blocking issues. The extension point is well-designed, backward compatible, and the motan LICENSE removal is verified-safe. Approving. The observations above (#1 and #2 are pre-existing and can be tracked separately) are optional follow-ups. See the two inline nits on the new interface for small API-doc improvements.