vk: fixup BLAS preallocation on Linux/amdgpu

Fixing pool allocator to properly signal allocation failure uncovered an
existing issue where we were lacking enough memory for dynamic model
BLASes on Linux/amdgpu. Erroneously the same memory region was used for
>1 BLAS. Surprisingly this hasn't led to any noticeable issues so far.

Increasing accels buffer size fixes the issue.

ref/vk/vk_ray_accel.c

@@ -80,7 +80,7 @@ static VkAccelerationStructureKHR createAccel(const char *name, VkAccelerationSt
 	const alo_block_t block = aloPoolAllocate(g_accel.accels_buffer_alloc, size, /*TODO why? align=*/256);
 
 	if (block.offset == ALO_ALLOC_FAILED) {
-		ERR("Failed to allocated %u bytes for blas \"%s\"", size, name);
+		ERR("Failed to allocate %u bytes for blas \"%s\"", size, name);
 		return VK_NULL_HANDLE;
 	}
 

ref/vk/vk_ray_internal.h

@@ -40,7 +40,7 @@ struct vk_combuf_s;
 qboolean createOrUpdateAccelerationStructure(struct vk_combuf_s *combuf, const as_build_args_t *args);
 
 #define MAX_SCRATCH_BUFFER (32*1024*1024)
-#define MAX_ACCELS_BUFFER (64*1024*1024)
+#define MAX_ACCELS_BUFFER (128*1024*1024)
 
 typedef struct {
 	// Geometry metadata. Lifetime is similar to geometry lifetime itself.