[TOPI][GPU] Mergepath sort with odd-even block sort (#7611)

* Mergepath sort with odd-even block sort

* fix lint, add test

* respond to review comments

* speed up tests by reducing dtype skews

* fix bad rebase

* change threading to support vulkan

* fix lint

* only sort if the data is non-empty

* fix lint again

* fix for vk

* move if to higher scope

* fix typo

Co-authored-by: Masahiro Masuda <masahi@[email protected]>

Author: Matthew Brookhart

include/tvm/tir/stmt.h

@@ -1312,6 +1312,10 @@ constexpr const char* fragment_shape = "fragment_shape";
  */
 constexpr const char* fragment_layout = "fragment_layout";
 
+/*!
+ * \brief Mark that the kernel is hand threaded and doesn't need syncs inserted
+ */
+constexpr const char* hand_threaded = "hand_threaded";
 /*!
  * \brief Check if attr_key is a pragma key extension
  * \param attr_key The attr key to be compared