feat(audit): M5.3 response-capture increments — request headers, ceiling-hits counter, per-method body opt-out (T7)

1. Request headers in Extra JSON (AuditWriteMiddleware): adds a `requestHeaders`
   object to the existing Extra JSON alongside remoteIp/userAgent; headers whose
   names appear in AuditLogOptions.HeaderRedactList (Authorization, X-Api-Key,
   Cookie, Set-Cookie by default) are replaced with "<redacted>" using
   OrdinalIgnoreCase matching — same policy as ScadaBridgeAuditRedactor.

2. AuditInboundCeilingHits counter: new IAuditInboundCeilingHitsCounter interface
   + NoOpAuditInboundCeilingHitsCounter default; AuditCentralHealthSnapshot
   implements the interface (Interlocked field, thread-safe) and exposes
   AuditInboundCeilingHits on IAuditCentralHealthSnapshot; AddAuditLog registers
   the NoOp default, AddAuditLogCentralMaintenance forwards to the snapshot;
   AuditWriteMiddleware accepts the counter as an optional ctor arg and increments
   it once per request where either the request or response body hit the cap.

3. Per-method SkipBodyCapture opt-out: adds SkipBodyCapture bool to
   PerTargetRedactionOverride; AuditWriteMiddleware consults the per-target
   override map at the start of InvokeAsync (before EnableBuffering) and, when
   set, skips body read + capture entirely — the audit row still emits with
   headers/metadata but null RequestSummary/ResponseSummary; truncation flags
   are also cleared so the ceiling-hits counter is not bumped for opted-out methods.

src/ZB.MOM.WW.ScadaBridge.AuditLog/Central/IAuditCentralHealthSnapshot.cs

@@ -50,6 +50,17 @@ public interface IAuditCentralHealthSnapshot
     /// </summary>
     int AuditRedactionFailure { get; }
 
+    /// <summary>
+    /// Count of inbound request/response body truncations at the
+    /// <see cref="ZB.MOM.WW.ScadaBridge.AuditLog.Configuration.AuditLogOptions.InboundMaxBytes"/>
+    /// ceiling since process start. Incremented by
+    /// <see cref="ZB.MOM.WW.ScadaBridge.InboundAPI.Middleware.AuditWriteMiddleware"/>
+    /// whenever either the request or response body exceeds the cap and is
+    /// truncated in the audit copy. A sustained non-zero count can indicate
+    /// callers sending unexpectedly large bodies.
+    /// </summary>
+    int AuditInboundCeilingHits { get; }
+
     /// <summary>
     /// Per-site latched stalled state: <c>true</c> when the
     /// <see cref="SiteAuditReconciliationActor"/> has observed two