Fixes crash of import memory with vkMemoryDedicatedAllocateInfo that have mismatched import handles

[leegao]

Commit 1

Currently, some drivers will hard-crash the dispatch->AllocateMemory (on older Mali drivers) if trying to import an image created with VK_EXTERNAL_MEMORY_HANDLE_TYPE_DMA_BUF_BIT_EXT into a:

   (VkExportMemoryAllocateInfo) {
      .sType = VK_STRUCTURE_TYPE_EXPORT_MEMORY_ALLOCATE_INFO,
      .pNext = pAllocateInfo->pNext,
      .handleTypes =
         VK_EXTERNAL_MEMORY_HANDLE_TYPE_ANDROID_HARDWARE_BUFFER_BIT_ANDROID,
   };

This PR introduces handle type tracking and validates the call to dispatch->AllocateMemory in the 3 handle import paths in wrapper_AllocateMemory to ensure that the wrong handle type won't crash the driver.

---

Commit 2

Fixes the blitting fallback path swapping blue/red on certain game engines (those preferring rgba8 swapchain)

Investigation: leegao/bionic-vulkan-wrapper#157 (comment)

The flow for a blit goes something like this:

GAME - renders to the primary image [r, g, b, a] or [b, g, r, a], physically backed by device local memory (so under TILING_OPTIMAL, configured in either RGBA or BGRA mode)
WSI - copies (byte-for-byte) the primary image to the secondary image, which is hardcoded as AHB_RGBA (since Mali doesn't support AHB_BGRA), but physically contains [b, g, r, a] memory for x11
X11 - imports the AHB_RGBA buffer containing assumed to be [b, g, r, a] data

this fails when the game picks [r, g, b, a] as the swapchain image, as the vkCopyImage will place [r, g, b, a] data instead of the expected [b, g, r, a] data into the AHB_RGBA secondary image (I know, this is confusing), hence swapping the b/r channels.

This commit reverts the part of 2af0e2e that introduced [r, g, b, a] as a valid swapchain format (originally meant as a fast path to avoid emulating bgra8)

---

Testing: see leegao/bionic-vulkan-wrapper#157 (comment) (still has to be done by thirdparty tester, many thanks to https://github.com/gtaforever00 and https://github.com/Artewar67)

For @utkarshdalal - would appreciate feeding this patch through code-review to double check everything as well

[leegao]

@utkarshdalal see 0bd55cb

the handle_types check is lifted up now, and we do an explicit unlink of the MemoryDedicatedAllocateInfo without an external import handle type