feat: System-AI cache surface (v0.8.0)

Picks up cache-related additions in HubAdmin:
- AskAi gains `forceFresh` request flag and `cached` /
  `cachedAt` / `cacheHits` response fields.
- New ClearSystemLlmCache + ForgetCachedExplanation RPCs.
- SystemAiStatus carries `cacheCount` so the UI can render
  "Cache: N cached explanations".

HubClient exposes `clearSystemLlmCache()` returning the
purged count and `forgetCachedExplanation(prompt)` for the
Regenerate flow.

Signed-off-by: flemming-it <stefan.a.flemming@googlemail.com>

lib/src/hub_client.dart

@@ -217,8 +217,31 @@ class HubClient {
   /// One-shot prompt to the configured System AI. Errors come
   /// back inside [AskAiResponse.errorKind] (not as exceptions)
   /// so Studio can map each category to its inline-fix copy.
-  Future<AskAiResponse> askAi(String prompt) {
-    return _admin.askAi(pb.AskAiRequest(prompt: prompt));
+  ///
+  /// Cache-aware: identical prompts hit the persistent
+  /// per-(model, privacy_mode, prompt) cache; the response
+  /// carries `cached: true` plus the original generation
+  /// latency. Pass [forceFresh] to skip the cache (Studio's
+  /// Regenerate button).
+  Future<AskAiResponse> askAi(String prompt, {bool forceFresh = false}) {
+    return _admin.askAi(
+      pb.AskAiRequest(prompt: prompt, forceFresh: forceFresh),
+    );
+  }
+
+  /// Drop every cached System-AI explanation. Used by Settings
+  /// when the operator wants a clean slate without changing
+  /// model/privacy mode (which would also flush).
+  Future<int> clearSystemLlmCache() async {
+    final r = await _admin.clearSystemLlmCache(Empty());
+    return r.purged.toInt();
+  }
+
+  /// Drop a single cached entry by prompt. Studio's Regenerate
+  /// button calls this so the next askAi falls through to the
+  /// live provider.
+  Future<void> forgetCachedExplanation(String prompt) async {
+    await _admin.forgetCachedExplanation(pb.AskAiRequest(prompt: prompt));
   }
 
   /// Persist a new System-AI configuration. Validates +