feat(v7-f54): cross-preset block transplant honesty gate

Closes V7-F54 (cppmega-mlx-rk4w): proves brick modularity — MoE
brick (from mixtral-style preset) grafted into a llama-style
attention→mlp chain still builds, trains, and surfaces routing
metrics.

Tests (tests/v4/test_block_transplant.py): 2/2 —
  * 3-brick graph [attention, moe, mlp] at H=128 trains 4 steps
    with finite losses + weight_delta > 0.
  * extras.moe still carries num_experts + top_k through the
    transplant.

Author: apstenku123

tests/v4/test_block_transplant.py

@@ -0,0 +1,69 @@
+"""V7-F54: cross-preset block transplant.
+
+Proves brick modularity: take MoE (from a mixtral-style preset),
+graft it into a llama-style attention→mlp chain, build, train.
+"""
+
+from __future__ import annotations
+
+import pytest
+
+from cppmega_v4.jsonrpc.schema import VerifyParams
+from cppmega_v4.runner import Pipeline, run_pipeline
+
+
+def _hybrid_spec() -> VerifyParams:
+    """attention → moe (transplanted) → mlp."""
+    return VerifyParams.model_validate({
+        "graph": {
+            "nodes": [
+                {"id": "attn", "kind": "attention",
+                 "params": {"num_heads": 4, "head_dim": 64}},
+                {"id": "moe_xplant", "kind": "moe",
+                 "params": {"num_experts": 4, "top_k": 2}},
+                {"id": "mlp", "kind": "mlp", "params": {}},
+            ],
+            "edges": [
+                {"src": "attn", "dst": "moe_xplant"},
+                {"src": "moe_xplant", "dst": "mlp"},
+            ],
+        },
+        "dim_env": {"B": 1, "S": 8, "H": 128,
+                    "nh": 2, "nkv": 1, "head_dim": 64,
+                    "num_experts": 4, "top_k": 2},
+        "loss": {"kind": "cross_entropy", "head_outputs": ["mlp"]},
+        "optim": {"kind": "adamw",
+                  "groups": [{"matcher": "all", "lr": 1e-3,
+                              "weight_decay": 0.01,
+                              "betas": [0.9, 0.95]}]},
+    })
+
+
+def test_v7_f54_transplanted_moe_trains_in_llama_chain():
+    rep = run_pipeline(_hybrid_spec(), Pipeline.from_dict({
+        "stages": ["parse", "verify_build_spec", "build_model",
+                   "train"],
+        "stage_options": {"train": {"num_steps": 4}},
+    }))
+    tr = next(s for s in rep.stages if s.name == "train")
+    assert tr.status == "ok", (
+        f"transplanted graph failed: {tr.error}"
+    )
+    for L in tr.extras["losses"]:
+        assert -1e10 < L < 1e10
+    assert tr.extras["weight_delta_norm"] > 0
+
+
+def test_v7_f54_moe_extras_surface_through_transplant():
+    """The transplanted MoE still surfaces extras.moe — its routing
+    metrics aren't lost just because it's in a foreign chain."""
+    rep = run_pipeline(_hybrid_spec(), Pipeline.from_dict({
+        "stages": ["parse", "verify_build_spec", "build_model",
+                   "train"],
+        "stage_options": {"train": {"num_steps": 2}},
+    }))
+    tr = next(s for s in rep.stages if s.name == "train")
+    moe = tr.extras["moe"]
+    assert moe is not None
+    assert moe["num_experts"] == 4
+    assert moe["top_k"] == 2