feat(srv6)!: split node information queries

Replace the combined node-info protocol with independent profile,
external-endpoint, and extra-route requests. Aggregate consumers use
NodeInfoQueryCoordinator and validate response source addresses.

BREAKING CHANGE: nodeinfotinyreq/resp and nodeinfofullreq/resp are
removed. Peers must use the profile, external-endpoint, and extra-route
request pairs.

AGENTS.md

@@ -1,54 +1,39 @@
 # KLALB Repository Guide
 
-KLALB is a Java SRv6/load-balancing system. The executable entrypoint is `org.kne.cloud.network.klalb.KLALBMain`; runtime configuration is `klalb-config.json` in the repository root.
+KLALB is a Java SRv6/load-balancing system. Start at `org.kne.cloud.network.klalb.KLALBMain`; runtime configuration is root-relative `klalb-config.json`.
 
 ## Build And Run
 
-- This is a plain Eclipse/IntelliJ Java project: sources are `src/`, vendored dependencies are `lib/`, and output is `bin/`. There is no Maven or Gradle.
-- `.classpath` targets `JavaSE-25`. Sources contain Chinese text, so manual compilation must use UTF-8:
+- This is an Eclipse Java project, not Maven or Gradle: `src/` contains sources, `lib/` vendored dependencies, and `bin/` compiled output. `.classpath` targets Java 25.
+- Source files contain Chinese text, so compile with UTF-8 from the repository root:
 
 ```powershell
 & "C:\Program Files\Zulu\zulu-25\bin\javac.exe" -encoding UTF-8 -cp "lib/*" -d bin (Get-ChildItem -Recurse src -Filter *.java | ForEach-Object FullName)
 ```
 
-- Run from the repository root. Manual `javac` does not copy resources, so keep `src` on the runtime classpath:
+- Run from the repository root. `src` must stay on the classpath because manual compilation does not copy resource bundles:
 
 ```powershell
 & "C:\Program Files\Zulu\zulu-25\bin\java.exe" --enable-native-access=ALL-UNNAMED "--add-opens=java.base/jdk.internal.misc=ALL-UNNAMED" -cp "bin;src;lib/*" org.kne.cloud.network.klalb.KLALBMain
 ```
 
-- The native libraries and `klalb-config.json` are resolved from the current directory. Restart a running JVM after recompiling.
-- TUN creation normally needs elevation. For non-admin UI/routing checks, set `"enableTUN": false`.
-- Current full compilation emits 11 pre-existing varargs/deprecation warnings; exit code `0` is success.
+- Native libraries and `klalb-config.json` are resolved from the current directory; restart the JVM after recompiling. TUN creation requires elevation, so use `"enableTUN": false` for non-admin checks.
 
 ## Dashboard
 
-- `dashboard/` is a Git submodule. Commit dashboard changes inside it, then update the parent repository's submodule pointer.
-- Run frontend commands from `dashboard/` with pnpm:
-
-```powershell
-pnpm install --frozen-lockfile
-pnpm lint
-pnpm typecheck
-pnpm build
-pnpm dev
-```
-
-- `pnpm build` runs `tsc -b` then Vite and writes `dashboard/dist`, which the Java web server hosts. Vite development proxies `/api` to `http://127.0.0.1:4665`.
-- Add shadcn components through `pnpm dlx shadcn@latest add <component>`; do not hand-create replacements for installed shadcn primitives.
+- `dashboard/` is a Git submodule. Commit dashboard changes in that repository, then update the parent submodule pointer.
+- Run frontend commands from `dashboard/`: `pnpm install --frozen-lockfile`, `pnpm lint`, `pnpm typecheck`, and `pnpm build`. The build is `tsc -b && vite build` and produces `dashboard/dist`, which the Java server hosts.
+- Vite proxies `/api` to `http://127.0.0.1:4665`; use `pnpm dev` only with the Java API running there.
 
 ## Verification
 
-- There is no CI or automated test suite. `*Test*` classes are manual harnesses that require real network peers.
-- For Java changes, compile and launch the app. For dashboard changes, run `pnpm typecheck` and `pnpm build`.
+- No repository test runner or CI workflow is configured. For Java changes, compile and launch the app; for dashboard changes, run `pnpm typecheck` and `pnpm build`.
 
 ## Important Boundaries
 
-- `KLALBConfigItem` is a polymorphic JSON array keyed by `Type`. Adding a type requires a subclass and cases in both default config serializer and deserializer; unknown types must remain preserved.
-- `/api/config` is field-by-field parsing, not whole-object Gson mapping. Keep legacy key aliases in sync with new fields.
-- Vendored Gson is `2.1`: HTTP responses that are `JsonElement` instances must be serialized with `JsonElement.toString()`, not reflective `gson.toJson(Object)`; configuration files must use the configured pretty-print Gson path rather than `JsonElement.toString()`.
+- `KLALBConfigItem` is a polymorphic JSON array keyed by case-sensitive `Type`. New types need serializer and deserializer support; preserve unknown items' raw JSON.
+- `/api/config` parses fields and legacy aliases explicitly. Web and Swing writes must use `KLALBProxySystem`'s revision-checked detached-candidate commit path, never mutate the canonical config directly.
+- Vendored Gson is 2.1: serialize HTTP `JsonElement` values with `toString()`; serialize configuration through the configured pretty-print Gson path.
 - UI strings use `UIEnv.getRsb()`; add keys to both `src/klalb_zh_CN.properties` and `src/klalb_en_US.properties`.
-- Web and Swing configuration writes must use `KLALBProxySystem`'s revisioned detached-candidate commit/event path; do not mutate the canonical config object directly.
-- `KLALBController.PublishedNodeInfo` is the thread-safe source for Tiny/Full node-info responses. Publish name, description, external endpoints, and Extra Routes through the controller method so snapshots and Tiny/Full update flags stay consistent.
-- `RouterInfo` no longer carries a device name. Its wire format retains an empty legacy UTF slot and `RouterInfoPacket` has optional Tiny/Full invalidation flags. Treat codec changes as compatibility work: preserve old-reader behavior and review a whole-mesh rollout.
-- Full node-info carries `extraRoutes` separately from the endpoint `data` list. Keep absent fields compatible with older peers.
+- `KLALBController.PublishedNodeInfo` is the thread-safe node-info source. Publish profile, effective external endpoints, and extra routes through the controller so snapshots and update flags remain coherent. Peer queries are separate profile, endpoint, and extra-route requests; use `NodeInfoQueryCoordinator` when a consumer needs an aggregate detail result.
+- `RouterInfo` retains an empty legacy UTF slot. `RouterInfoPacket` appends optional update flags behind a marker; treat binary codec changes as mesh-compatibility work and preserve old-reader behavior.