[Diffusion] Add native KV cache initialization and Scheduler-managed block allocation (#6094)

Signed-off-by: zwhzzz0821 <zwhzzz0821@users.noreply.github.com>
Co-authored-by: zwhzzz0821 <zwhzzz0821@users.noreply.github.com>

tests/diffusion/diffusion_kv/test_request.py

@@ -107,11 +107,8 @@ def test_request_rejects_non_context_values() -> None:
 def _manager(*, enable_caching: bool) -> KVCacheManager:
     # vLLM 0.26 lazily registers its built-in KV specs. Another test may have
     # already registered an out-of-tree spec, making the registry non-empty
-    # before FullAttentionSpec is registered. Initialize the native mappings
-    # explicitly; vLLM 0.25 has no registration hook and needs no setup.
-    register_all = getattr(native_kv_managers, "register_all_kvcache_specs", None)
-    if register_all is not None:
-        register_all(None)
+    # before FullAttentionSpec is registered. Initialize the native mappings.
+    native_kv_managers.register_all_kvcache_specs(None)
 
     spec = FullAttentionSpec(
         block_size=BLOCK_SIZE,