BLI: reduce wasted memory in linear allocator

The main change is that large allocations are done separately now.
Also, buffers that small allocations are packed into, have a maximum
size now. Using larger buffers does not really provider performance
benefits, but increases wasted memory.

source/blender/blenlib/tests/BLI_linear_allocator_test.cc

@@ -1,6 +1,7 @@
 /* Apache License, Version 2.0 */
 
 #include "BLI_linear_allocator.hh"
+#include "BLI_rand.hh"
 #include "BLI_strict_flags.h"
 #include "testing/testing.h"
 
@@ -115,4 +116,24 @@ TEST(linear_allocator, ConstructArrayCopy)
   EXPECT_EQ(span2[2], 3);
 }
 
+TEST(linear_allocator, AllocateLarge)
+{
+  LinearAllocator<> allocator;
+  void *buffer1 = allocator.allocate(1024 * 1024, 8);
+  void *buffer2 = allocator.allocate(1024 * 1024, 8);
+  EXPECT_NE(buffer1, buffer2);
+}
+
+TEST(linear_allocator, ManyAllocations)
+{
+  LinearAllocator<> allocator;
+  RandomNumberGenerator rng;
+  for (int i = 0; i < 1000; i++) {
+    int size = rng.get_int32(10000);
+    int alignment = 1 << (rng.get_int32(7));
+    void *buffer = allocator.allocate(size, alignment);
+    EXPECT_NE(buffer, nullptr);
+  }
+}
+
 }  // namespace blender::tests