[Perf] (MG_State/FastSTL): make object deletion cheap again

- GLContext::MarkBufferObjectForDeletion now detaches the deleted buffer
  only from the currently bound VAO (GL 4.6 5.1.2 semantics; other VAOs
  keep their shared_ptr attachments alive). The old every-VAO scan was
  O(VAOs) per delete - with one VAO per chunk section, vanilla chunk
  churn made it dominate the render thread and FPS decay over minutes.
- Bump FastSTL: erase(key) destroys in place instead of building the
  discarded successor iterator (a linear bucket-array scan), and switch
  the buffer/framebuffer/renderbuffer deletion paths to the key overload.

Together these removed the 34% render-thread deletion overhead measured
in aged vanilla sessions (simpleperf, Adreno 830 / DirectVulkan).

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

MobileGL/MG_State/GLState/BufferState/BufferState.cpp

@@ -64,7 +64,9 @@ namespace MobileGL::MG_State::GLState {
                         }
                     }
                 }
-                m_bufferObjects.erase(it);
+                // Key-based erase skips FastSTL's successor-iterator scan, which is
+                // pure overhead here and dominates delete-heavy frames.
+                m_bufferObjects.erase(index);
             }
             m_indexGenerator.Delete(index);
         }

MobileGL/MG_State/GLState/Core.cpp

@@ -80,11 +80,16 @@ namespace MobileGL::MG_State {
 
         void GLContext::MarkBufferObjectForDeletion(Uint index) {
             if (ValidateBufferObject(index)) {
+                // GL semantics: deleting a buffer detaches it only from the CURRENT
+                // context's bindings, including the currently bound VAO's attachment
+                // points; attachments in other VAOs must survive (the shared_ptr keeps
+                // the data object alive, matching the spec's deferred deletion). The
+                // previous every-VAO scan was wrong per spec and O(VAOs) per delete —
+                // with one VAO per chunk section, vanilla's steady buffer churn made it
+                // dominate the render thread and FPS decay over session time.
                 auto bufferObject = m_bufferState.GetBufferObject(index);
-                auto& vaos = m_vertexArrayState.GetAllVertexArrays();
-                for (auto& vao : vaos) {
-                    if (vao == nullptr) continue;
-
+                const auto& vao = m_vertexArrayState.GetBoundVertexArray();
+                if (vao != nullptr) {
                     if (vao->GetIndexBufferBindingSlot().GetBoundObject() == bufferObject) {
                         vao->GetIndexBufferBindingSlot().Bind(nullptr);
                     }

MobileGL/MG_State/GLState/FramebufferState/FramebufferState.cpp

@@ -65,7 +65,7 @@ namespace MobileGL::MG_State::GLState {
                         bindingSlot.Bind(GetFramebufferObject(0));
                     }
                 }
-                m_framebufferObjects.erase(it);
+                m_framebufferObjects.erase(index);
             }
             m_indexGenerator.Delete(index);
         }

MobileGL/MG_State/GLState/RenderbufferState/RenderbufferState.cpp

@@ -57,7 +57,7 @@ namespace MobileGL::MG_State::GLState {
                         bindingSlot.Bind(nullptr);
                     }
                 }
-                m_renderbufferObjects.erase(it);
+                m_renderbufferObjects.erase(index);
             }
             m_indexGenerator.Delete(index);
         }