[BUG] Sync-data-center: no queues, no batching, no dedup on the gateway side

## Description
There are no queues and no batching anywhere in the gateway-side sync path. A `MYSELF`/`REFRESH` payload containing the entire namespace deserializes and re-caches every entry inline on the `onMessage` thread. No event dedup: `AbstractDataHandler.handle` always re-applies (`doUpdate`/`doRefresh`) regardless of whether the data actually changed — admin churn translates 1:1 into gateway CPU. A slow plugin handler stalls all subsequent config updates for that gateway.

## Location
```
shenyu-sync-data-websocket/.../client/ShenyuWebsocketClient.java:189-207,267-274
WebsocketDataHandler.java:67 -> AbstractDataHandler.java:62-85
AbstractNodeDataSyncService.java:212-253
```

## Impact
A single slow handler or a large burst saturates the one listener thread and silently delays all later config propagation; admin churn amplifies 1:1 into gateway CPU.

## Suggested fix
Add a bounded work queue + parallelism for config application; dedup by version/md5 before re-applying.

## Related existing issue(s)
#6461 (websocket reconnect) and #6523 (repeated successful polls) are specific defects; this is the structural lack of gateway-side queueing/dedup that amplifies them.

_Identified during the 2026-08-02 audit; full list in [`docs/issue-candidates-2026-08-02.md`](docs/issue-candidates-2026-08-02.md)._

Start with ShenyuWebsocketClient.java:189-207,267-274 and trace WebsocketDataHandler.java:67 into AbstractDataHandler.java:62-85 and AbstractNodeDataSyncService.java:212-253. Determine how message handling currently applies updates and refreshes inline, then define bounded queueing, parallel application, and version/md5 deduplication; done means slow handlers and repeated unchanged events no longer block or reapply gateway configuration.

Written by the indexing model from the issue text.