fix: address code quality review findings for batch 1

- SubjectsCollide: split tokens once upfront instead of O(n²) TokenAt calls
- NatsHeaderParser: manual digit accumulation avoids string allocation and overflow
- NatsHeaders: use IReadOnlyDictionary for Headers, immutable Invalid sentinel
- PermissionLruCache: add missing Count property

src/NATS.Server/Subscriptions/SubjectMatch.cs

@@ -172,11 +172,13 @@ public static class SubjectMatch
         if (lit2 && !lit1)
             return MatchLiteral(subj2, subj1);
 
-        // Both have wildcards
-        int n1 = NumTokens(subj1);
-        int n2 = NumTokens(subj2);
-        bool hasFwc1 = subj1.Contains('>');
-        bool hasFwc2 = subj2.Contains('>');
+        // Both have wildcards — split once to avoid O(n²) TokenAt calls
+        var tokens1 = subj1.Split(Sep);
+        var tokens2 = subj2.Split(Sep);
+        int n1 = tokens1.Length;
+        int n2 = tokens2.Length;
+        bool hasFwc1 = tokens1[^1] == ">";
+        bool hasFwc2 = tokens2[^1] == ">";
 
         if (!hasFwc1 && !hasFwc2 && n1 != n2)
             return false;
@@ -188,9 +190,7 @@ public static class SubjectMatch
         int stop = Math.Min(n1, n2);
         for (int i = 0; i < stop; i++)
         {
-            var t1 = TokenAt(subj1, i);
-            var t2 = TokenAt(subj2, i);
-            if (!TokensCanMatch(t1, t2))
+            if (!TokensCanMatch(tokens1[i], tokens2[i]))
                 return false;
         }
         return true;