fix(deepseek): comment out shm_size (redundant with --privileged)

recipes/deepseek-v4-flash/recipe.toml

@@ -7,9 +7,8 @@ tensor_parallel_size = 2
 pipeline_parallel_size = 1
 distributed_backend = "mp"
 
-# DeepSeek V4 with 1M context needs much more shared memory than the
-# default 10g.  KV cache + activations for 1M tokens fills up fast.
-shm_size = "64gb"
+# --privileged makes --shm-size redundant; kept for documentation
+# shm_size = "64gb"
 
 entrypoint = ["vllm", "serve"]