chore(profiling): instrument plugin-load hot paths (v1.4.9 R3)

Bump AutoTranslate-warmup and FilterAllTabs log-level from Debug to
Information so the xllog tail surfaces them without a Debug filter.
Wrap MessageStore.Connect and MessageStore.Migrate in Stopwatches so
the SQLite open and migration-chain costs are visible too.

Sub-Task 3.4 Befund on v1.4.8-baseline (4 reloads, medians):
- MessageStore.Connect: 50.5 ms
- MessageStore.Migrate:    2 ms
- MessageManager.FilterAllTabs: 68.5 ms
- AutoTranslate warmup:  108 ms
- UiBuilder HITCH:       108.9 ms

Outcome D — none of the three dominates the 200 ms threshold. The
ChatTwo "300 ms" comment for AutoTranslate is falsified at ~108 ms;
SQLite is not the bottleneck (52.5 ms total); FilterAllTabs runs on
the worker thread and only competes for CPU slots. The HITCH is left
unexplained by these probes, which keeps Hypothesis c (multi-window
WindowSystem.Draw initial pass) as the main R2 suspect to be
validated by the R1 lazy-window refactor.

Logs stay in as belt-and-suspenders for future plugin-load
regressions.

HellionChat/MessageManager.cs

@@ -206,7 +206,10 @@ internal class MessageManager : IAsyncDisposable
                 Plugin.LogProxy.Error(ex, "Error in FilterAllTabs");
             }
 
-            Plugin.LogProxy.Debug($"FilterAllTabs took {stopwatch.ElapsedMilliseconds}ms");
+            // v1.4.9 R3 profiling: Information so the xllog tail surfaces this
+            // without a Debug filter. Belt-and-suspenders for future plugin-load
+            // regressions; remains in place after Sub-Task 3.4 Befund.
+            Plugin.LogProxy.Information($"FilterAllTabs took {stopwatch.ElapsedMilliseconds}ms");
         });
     }
 

HellionChat/MessageStore.cs

@@ -237,14 +237,26 @@ internal class MessageStore : IDisposable
 
     private SqliteConnection Connect()
     {
+        // v1.4.9 R3 profiling: trace cost of SQLite open + pragma-apply. Paired
+        // with the Migrate-Stopwatch below — Connect alone is the cheap half
+        // (Open + a handful of PRAGMAs); the expensive half typically lives in
+        // Migrate, especially on a large DB after a schema bump.
+        var connectSw = System.Diagnostics.Stopwatch.StartNew();
         var conn = new SqliteConnection(BuildConnectionString(DbPath));
         conn.Open();
         ApplyPragmas(conn);
+        connectSw.Stop();
+        _logger.Information($"MessageStore.Connect took {connectSw.ElapsedMilliseconds}ms");
         return conn;
     }
 
     private void Migrate()
     {
+        // v1.4.9 R3 profiling: trace cost of the schema-migration chain. On a
+        // large DB after a fresh schema bump this is the dominant SQLite cost
+        // at plugin-load, not Connect.
+        var migrateSw = System.Diagnostics.Stopwatch.StartNew();
+
         using var cmd = Connection.CreateCommand();
         cmd.CommandText = "PRAGMA user_version;";
         var userVersion = Convert.ToInt32(cmd.ExecuteScalar());
@@ -276,6 +288,9 @@ internal class MessageStore : IDisposable
 
         foreach (var migration in migrationsToDo)
             migration();
+
+        migrateSw.Stop();
+        _logger.Information($"MessageStore.Migrate took {migrateSw.ElapsedMilliseconds}ms");
     }
 
     private void Migrate0()

HellionChat/Util/AutoTranslate.cs

@@ -62,7 +62,12 @@ internal static class AutoTranslate
         {
             var sw = Stopwatch.StartNew();
             AllEntries();
-            Plugin.LogProxy.Debug($"Warming up auto-translate took {sw.ElapsedMilliseconds}ms");
+            // v1.4.9 R3 profiling: Information so the xllog tail surfaces this
+            // without a Debug filter. Belt-and-suspenders for future plugin-load
+            // regressions; remains in place after Sub-Task 3.4 Befund.
+            Plugin.LogProxy.Information(
+                $"Warming up auto-translate took {sw.ElapsedMilliseconds}ms"
+            );
         })
         {
             IsBackground = true,