fix(tests): migrate the shared test database once per session

The four WebApplicationFactory variants are each their own ClassDataSource,
so each boots its own API host, but all of them share a single
PerTestSession Postgres container. Every host applied migrations on startup,
and EF's migrator is not safe to run concurrently: two hosts booting
together both see the same pending migrations and both apply them.

TUnit 1.70 happened to schedule those initializations so the first migrator
finished before the others started; 1.74 does not. That surfaced the latent
race as 30 failures in API.IntegrationTests - "type/relation/collation
already exists", duplicate keys on __EFMigrationsHistory, and queries
hitting a half-migrated schema.

Apply migrations in InMemoryDatabase.InitializeAsync, which runs once per
session, and set SKIPMIGRATION on the factories so no host migrates on boot.
This mirrors what Cron.IntegrationTests already does for the Cron host. It
also removes the stale Npgsql type-catalog hazard, since the schema is
complete before any host opens its first connection, and runs the migrations
once rather than up to four times.

Author: hhvrc

API.IntegrationTests/Docker/InMemoryDatabase.cs

@@ -1,4 +1,7 @@
-using OpenShock.Common.Utils;
+using Microsoft.EntityFrameworkCore;
+using Microsoft.Extensions.Logging.Abstractions;
+using OpenShock.Common.OpenShockDb;
+using OpenShock.Common.Utils;
 using Testcontainers.PostgreSql;
 using TUnit.Core.Interfaces;
 
@@ -27,7 +30,25 @@ public PostgreSqlContainer Container
             return _container;
         }
     }
-    
-    public Task InitializeAsync() => Container.StartAsync();
+
+    /// <summary>
+    /// Starts the container and brings the schema up to date. Migrations are applied here, once per
+    /// test session, rather than by the API hosts on boot: several <see cref="WebApplicationFactory"/>
+    /// variants share this one database, and each boots its own host. EF's migrator is not safe to run
+    /// concurrently against the same database, so letting each host migrate on startup races - two
+    /// migrators both see the same pending migrations and both apply them, which fails with
+    /// "type/relation/collation already exists", a duplicate key on __EFMigrationsHistory, or queries
+    /// hitting a half-migrated schema. The factories set OPENSHOCK__DB__SKIPMIGRATION so this is the
+    /// only migrator, matching how Cron.IntegrationTests already seeds its schema.
+    /// </summary>
+    public async Task InitializeAsync()
+    {
+        await Container.StartAsync();
+
+        await using var migrationContext = new MigrationOpenShockContext(
+            Container.GetConnectionString(), false, NullLoggerFactory.Instance);
+        await migrationContext.Database.MigrateAsync();
+    }
+
     public ValueTask DisposeAsync() => Container.DisposeAsync();
 }

API.IntegrationTests/WebApplicationFactory.cs

@@ -51,7 +51,10 @@ protected override void ConfigureWebHost(IWebHostBuilder builder)
             { "ASPNETCORE_UNDER_INTEGRATION_TEST", "1" },
 
             { "OPENSHOCK__DB__CONN", PostgreSql.Container.GetConnectionString() },
-            { "OPENSHOCK__DB__SKIPMIGRATION", "false" },
+            // InMemoryDatabase migrates the shared database once per test session. Every factory
+            // variant boots its own host against that same database, and concurrent EF migrators
+            // corrupt each other, so no host migrates on boot here.
+            { "OPENSHOCK__DB__SKIPMIGRATION", "true" },
             { "OPENSHOCK__DB__DEBUG", "false" },
 
             { "OPENSHOCK__REDIS__CONN", Redis.Container.GetConnectionString() },