config: route parallel config-leaf reads through get_parallel() (#33170)

The parallel namespace joins the accessor migration: 106 config-leaf reads
(enable_dp_lm_head, enable_dp_attention, pp_async_batch_depth, dp_size,
ep_join_rank_offset, dwdp_size, ...) flip from get_server_args()/
self.server_args to get_parallel(), which serves config leaves from the
published parallel bag via __getattr__.

- ParallelContext.__getattr__ is restructured to stay dynamo-traceable
  (object.__getattribute__ graph-breaks): gate helpers such as
  enable_moe_dense_fully_dp() run inside compiled model forwards. A
  fullgraph regression test pins the pattern.
- The five live-shadowed topology sizes (tp/pp/dcp/attn_cp/moe_dp_size)
  keep their server_args reads: the live @property wins on the accessor,
  and conditionally-initialized groups would fail loud at unconditional
  call sites.
- Elastic-EP scale writers (ep_size/dp_size x4 in model_runner) reroute
  to get_context().override together with their remaining instance
  readers (expert_location gpus-per-node paths); the ServerArgs.override
  ratchet drops 39 -> 35.
- The expert placement helpers (compute_logical_to_rank_dispatch_
  physical_map, _compute_logical_to_all_physical_map,
  _prefer_same_node_experts) now read everything from the bags and drop
  their server_args parameter; their unit tests publish the config they
  need instead of stubbing it.

python/sglang/srt/runtime_context.py

@@ -125,12 +125,19 @@ class ParallelContext:
 
     def __getattr__(self, name):
         # Reached only for names that are neither a live @property nor a slot:
-        # serve parallel config leaves from the published bag.
-        try:
-            config = object.__getattribute__(self, "_config")
-        except AttributeError:
-            config = None
-        if config is not None and name in config:
+        # serve parallel config leaves from the published bag. The body must
+        # stay dynamo-traceable — config-leaf reads such as
+        # ``get_parallel().moe_dense_tp_size`` run inside compiled model
+        # forwards, and ``object.__getattribute__`` graph-breaks.
+        if name.startswith("_"):
+            # No config leaf is underscored; this also breaks the recursion
+            # when the ``_config`` slot itself is still unset (pickle/copy
+            # protocols probe attributes before __init__ runs).
+            raise AttributeError(name)
+        config = self._config
+        # ``_fields`` is a plain ``__dict__`` entry on the bag; ``in`` on the
+        # dict avoids ``_ConfigBag.__contains__`` (not traceable).
+        if config is not None and name in config._fields:
             return getattr(config, name)
         detail = (
             "not a published parallel config leaf"