mlx: drop carried metal residency patch now that it is upstream (#18854)

MLX_VERSION

@@ -1 +1 @@
-264c14fe650e1dd9bb0c3902dc06b08033d8f89c
+a59cc23190f1e26fabcba8693607bb4c34b2ac2b

mlx/compat/mlx/0001-metal-residency-refresh.patch

@@ -1,84 +0,0 @@
-diff --git a/mlx/backend/metal/resident.cpp b/mlx/backend/metal/resident.cpp
---- a/mlx/backend/metal/resident.cpp
-+++ b/mlx/backend/metal/resident.cpp
-@@ -1,7 +1,8 @@
--// Copyright © 2024 Apple Inc.
-+// Copyright © 2024-2026 Apple Inc.
- 
- #include <algorithm>
- #include <cassert>
-+#include <chrono>
- #include <cstdio>
- #include <sstream>
- #include <stdexcept>
-@@ -43,10 +44,38 @@
-       }
-       throw std::runtime_error(msg.str());
-     }
--  }
--}
--
--ResidencySets::~ResidencySets() = default;
-+    if (int interval =
-+            env::get_var("MLX_METAL_RESIDENCY_REFRESH_INTERVAL_MS", 0);
-+        interval > 0) {
-+      refresh_thread_ = std::thread([this, interval] {
-+        std::unique_lock<std::mutex> lock(mtx_);
-+        while (!refresh_cv_.wait_for(
-+            lock, std::chrono::milliseconds(interval), [this] {
-+              return stop_refresh_;
-+            })) {
-+          auto pool = new_scoped_memory_pool();
-+          // macOS can drop residency after GPU idle despite a standing request.
-+          for (auto& s : sets_) {
-+            if (s.size != 0) {
-+              s.set->requestResidency();
-+            }
-+          }
-+        }
-+      });
-+    }
-+  }
-+}
-+
-+ResidencySets::~ResidencySets() {
-+  if (refresh_thread_.joinable()) {
-+    {
-+      std::lock_guard<std::mutex> lock(mtx_);
-+      stop_refresh_ = true;
-+    }
-+    refresh_cv_.notify_one();
-+    refresh_thread_.join();
-+  }
-+}
- 
- bool ResidencySets::add_set_locked(std::string* error_out) {
-   NS::SharedPtr<MTL::ResidencySet> set;
-diff --git a/mlx/backend/metal/resident.h b/mlx/backend/metal/resident.h
---- a/mlx/backend/metal/resident.h
-+++ b/mlx/backend/metal/resident.h
-@@ -1,11 +1,13 @@
--// Copyright © 2024 Apple Inc.
-+// Copyright © 2024-2026 Apple Inc.
- 
- #pragma once
- 
- #include <atomic>
-+#include <condition_variable>
- #include <cstdint>
- #include <mutex>
- #include <string>
-+#include <thread>
- #include <unordered_map>
- #include <vector>
- 
-@@ -119,6 +121,9 @@
-   // taking the lock.
-   std::atomic<uint64_t> num_sets_{0};
-   mutable std::mutex mtx_;
-+  std::condition_variable refresh_cv_;
-+  bool stop_refresh_{false};
-+  std::thread refresh_thread_;
- };
- 
- } // namespace mlx::core::metal