buffer_cache: Skip cache on small uploads on Vulkan

Ports from OpenGL the optimization to skip small 3D uniform buffer
uploads. This will take advantage of the previously introduced stream
buffer.

Fixes instances where the staging buffer offset was being ignored.

src/video_core/renderer_vulkan/vk_buffer_cache.h

@@ -69,6 +69,13 @@ public:
 
     void BindTransformFeedbackBuffer(u32 index, VkBuffer buffer, u32 offset, u32 size);
 
+    std::span<u8> BindMappedUniformBuffer([[maybe_unused]] size_t stage,
+                                          [[maybe_unused]] u32 binding_index, u32 size) {
+        const StagingBufferRef ref = staging_pool.Request(size, MemoryUsage::Upload);
+        BindBuffer(ref.buffer, static_cast<u32>(ref.offset), size);
+        return ref.mapped_span;
+    }
+
     void BindUniformBuffer(VkBuffer buffer, u32 offset, u32 size) {
         BindBuffer(buffer, offset, size);
     }

src/video_core/renderer_vulkan/vk_staging_buffer_pool.cpp

@@ -86,7 +86,8 @@ StagingBufferPool::StagingBufferPool(const Device& device_, MemoryAllocator& mem
         .pNext = nullptr,
         .flags = 0,
         .size = STREAM_BUFFER_SIZE,
-        .usage = VK_BUFFER_USAGE_TRANSFER_SRC_BIT | VK_BUFFER_USAGE_INDEX_BUFFER_BIT,
+        .usage = VK_BUFFER_USAGE_TRANSFER_SRC_BIT | VK_BUFFER_USAGE_UNIFORM_BUFFER_BIT |
+                 VK_BUFFER_USAGE_INDEX_BUFFER_BIT,
         .sharingMode = VK_SHARING_MODE_EXCLUSIVE,
         .queueFamilyIndexCount = 0,
         .pQueueFamilyIndices = nullptr,