[YDF] Faster OOB computation

Cherry-pick from google#324. Replaces single
mutex OOB computation with striped locking and an OOB gate mechanism for
better multi-threaded performance.

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

Author: ariellubonja

yggdrasil_decision_forests/learner/random_forest/BUILD

@@ -55,13 +55,15 @@ cc_library_ydf(
         "//yggdrasil_decision_forests/utils:synchronization_primitives",
         "//yggdrasil_decision_forests/utils:time",
         "//yggdrasil_decision_forests/utils:parallel_chrono",
+        "@com_google_absl//absl/cleanup",
         "@com_google_absl//absl/container:flat_hash_set",
         "@com_google_absl//absl/log",
         "@com_google_absl//absl/status",
         "@com_google_absl//absl/status:statusor",
         "@com_google_absl//absl/strings",
         "@com_google_absl//absl/strings:str_format",
         "@com_google_absl//absl/time",
+        "@com_google_absl//absl/types:span",
     ],
     alwayslink = 1,
 )
@@ -126,6 +128,7 @@ cc_test(
         "@com_google_absl//absl/status:statusor",
         "@com_google_absl//absl/strings",
         "@com_google_absl//absl/time",
+        "@com_google_absl//absl/types:span",
         "@com_google_googletest//:gtest_main",
     ],
 )