feat(studio): cached/Regenerate UX in audit Explain (v0.16.0)

Audit drilldown's explanation panel now surfaces whether the
answer came from the cache:
- "cached" pill next to the SYSTEM AI label, tooltip shows
  the original generation date + hit count.
- Latency renders as "orig 47000 ms" so the operator sees
  what a live call would have cost.
- Refresh icon next to the latency triggers Regenerate —
  drops the entry, re-asks the model.

System-AI editor shows a cache-status row when the cache is
non-empty with a one-click Clear button. Hidden on fresh
installs to keep the dialog quiet.

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

lib/data/hub.dart

@@ -119,6 +119,7 @@ class HubService {
       model: r.model,
       privacyMode: r.privacyMode,
       apiKeyEnv: r.apiKeyEnv,
+      cacheCount: r.cacheCount.toInt(),
     );
   }
 
@@ -126,15 +127,29 @@ class HubService {
   /// answer + latency on success or an [AskAiResult] with
   /// `errorKind` set on failure (never throws). Studio maps
   /// the error kind to its inline-fix copy.
-  Future<AskAiResult> askAi(String prompt) async {
-    final r = await _client.askAi(prompt);
+  Future<AskAiResult> askAi(String prompt, {bool forceFresh = false}) async {
+    final r = await _client.askAi(prompt, forceFresh: forceFresh);
     return AskAiResult(
       errorKind: r.errorKind,
       text: r.text,
       latencyMs: r.latencyMs,
+      cached: r.cached,
+      cachedAt: r.cachedAt,
+      cacheHits: r.cacheHits,
     );
   }
 
+  /// Drop every cached System-AI explanation. Returns the count
+  /// of entries that were purged.
+  Future<int> clearSystemLlmCache() {
+    return _client.clearSystemLlmCache();
+  }
+
+  /// Forget a single cached entry by prompt (for "Regenerate").
+  Future<void> forgetCachedExplanation(String prompt) {
+    return _client.forgetCachedExplanation(prompt);
+  }
+
   /// Persist a new System-AI configuration. The hub writes back
   /// to `~/.fai/config.yaml` and hot-reloads the in-memory copy
   /// in one round-trip; the caller gets the resulting status
@@ -160,6 +175,7 @@ class HubService {
       model: r.model,
       privacyMode: r.privacyMode,
       apiKeyEnv: r.apiKeyEnv,
+      cacheCount: r.cacheCount.toInt(),
     );
   }
 
@@ -687,6 +703,10 @@ class SystemAiStatus {
   /// off / redacted / full.
   final String privacyMode;
   final String apiKeyEnv;
+  /// Number of cached System-AI explanations currently stored.
+  /// Surfaced in Doctor + Settings so operators see the cache
+  /// earning its keep.
+  final int cacheCount;
 
   const SystemAiStatus({
     required this.enabled,
@@ -695,6 +715,7 @@ class SystemAiStatus {
     required this.model,
     required this.privacyMode,
     required this.apiKeyEnv,
+    this.cacheCount = 0,
   });
 }
 
@@ -705,13 +726,25 @@ class AskAiResult {
   /// On success: the assistant's reply.
   /// On failure: a human-readable diagnostic.
   final String text;
-  /// Round-trip latency in ms (success only).
+  /// Round-trip latency in ms (success only). For cache hits
+  /// this is the *original* generation latency.
   final int latencyMs;
+  /// True iff the answer came from the persistent cache.
+  final bool cached;
+  /// ISO-8601 timestamp the cached entry was first generated.
+  /// Empty for fresh answers.
+  final String cachedAt;
+  /// How many times this entry has been served from cache.
+  /// 0 for fresh answers.
+  final int cacheHits;
 
   const AskAiResult({
     required this.errorKind,
     required this.text,
     required this.latencyMs,
+    this.cached = false,
+    this.cachedAt = '',
+    this.cacheHits = 0,
   });
 
   bool get isSuccess => errorKind.isEmpty;