feat(pool-manager): rework stuck-runner cleanup to eliminate OOMs and cut DB load

### Why
* A production incident (`System.OutOfMemoryException` during `CheckForStuckRunners`) showed that we were
  loading **all** runners and their full `Lifecycle` collections into memory, then letting EF Core’s
  change-tracker churn on tens of thousands of entities.
* Each cleanup cycle recreated that pressure, eventually crashing the autoscaler pod and stopping the host.

### What
* **Query trimmed to SQL only**
  * Replaced `Include(...).AsEnumerable().Where(...)` with a *pure* LINQ-to-Entities
    filter:
    `LastState == Created && CreatedTime < now-10min`.
  * Removed unconditional `Include(x => x.Lifecycle)`.
  * Added `AsNoTracking()` and a **projection** to an anonymous type
    (`Select(r ⇒ new { … })`) so no full `Runner` entities are tracked.
* **Context lifetime & tracking**
  * Scoped `ActionsRunnerContext` with `await using` – guarantees disposal after
    the method exits.
  * Disabled auto-detection of changes only for this batch
    (`ChangeTracker.AutoDetectChangesEnabled = false`) to minimise
    change-tracker work.
* **Batch insert of lifecycle events**
  * Accumulate new `RunnerLifecycle` rows in an in-memory list and call
    `AddRange` once instead of adding per runner.
  * Single `SaveChangesAsync()` at the end → one DB round-trip.
* **Queue deletion without materialising collections**
  * For every stuck runner enqueue a `DeleteRunnerTask` directly using the
    projected key data (no need for full entity).
* **Logging**
  * Added explicit warning log per stuck runner **and** kept the original message
    structure for easy grepping.

### Result
* `CheckForStuckRunners` now pulls only the small “actually stuck” set into
  memory and never tracks existing `Lifecycle` rows.

**IN CLAUDE WE TRUST**

Author: gcordalis

PoolManager.cs

@@ -146,34 +146,50 @@ private async Task CheckForStuckRunners(List<GithubTargetConfiguration> targetCo
     {
         // check the database for runners that are in "created" state for more then 5 minutes.
 
-        var db = new ActionsRunnerContext();
-        foreach(var stuckRunner in db.Runners.Include(x => x.Lifecycle).AsEnumerable().Where(x => x.LastState == RunnerStatus.Created))
+        await using var db = new ActionsRunnerContext();
+        var cutoffTime = DateTime.UtcNow - TimeSpan.FromMinutes(10);
+        
+        // Query stuck runners without loading lifecycle collections
+        var stuckRunners = await db.Runners
+            .AsNoTracking()
+            .Where(x => x.LastState == RunnerStatus.Created && x.CreatedTime < cutoffTime)
+            .Select(x => new { x.RunnerId, x.CloudServerId, x.Hostname, x.Cloud })
+            .ToListAsync();
+        
+        if (stuckRunners.Count == 0)
+            return;
+        
+        // Process stuck runners and create lifecycle entries
+        var lifecycleEntries = new List<RunnerLifecycle>();
+        
+        foreach(var stuckRunner in stuckRunners)
         {
-
-            // check if runner is old enough to be stuck
-            if (stuckRunner.CreatedTime + TimeSpan.FromMinutes(10) > DateTime.UtcNow)
-                continue;
+            // Add to deletion queue
+            _queues.DeleteTasks.Enqueue(new DeleteRunnerTask
+            {
+                ServerId = stuckRunner.CloudServerId,
+                RunnerDbId = stuckRunner.RunnerId
+            });
 
-            // Note stuckness in lifecycle and add runner to deletion queue
-            stuckRunner.Lifecycle.Add(new RunnerLifecycle
+            // Create lifecycle entry for batch insert
+            lifecycleEntries.Add(new RunnerLifecycle
             {
+                RunnerId = stuckRunner.RunnerId,
                 Event = "Stuck in provisioning. Killing.",
                 EventTimeUtc = DateTime.UtcNow,
                 Status = RunnerStatus.Failure
             });
-           
-            _queues.DeleteTasks.Enqueue(new DeleteRunnerTask
-            {
-                ServerId = stuckRunner.CloudServerId,
-                RunnerDbId = stuckRunner.RunnerId
-            });
 
             _logger.LogWarning($"Killing Runner stuck in provisioning: {stuckRunner.Hostname} on {stuckRunner.Cloud}");
-            
         }
 
-        // write to DB
-        await db.SaveChangesAsync();
+        // Batch insert lifecycle entries without change tracking
+        if (lifecycleEntries.Count > 0)
+        {
+            db.ChangeTracker.AutoDetectChangesEnabled = false;
+            db.RunnerLifecycles.AddRange(lifecycleEntries);
+            await db.SaveChangesAsync();
+        }
     }
 
     private async Task ProcessStats(List<GithubTargetConfiguration> targetConfig)