doc: document MTP attention requirement for higher acceptance

The MTP head has attention weights (Q/K/V) but they are currently unused
(FFN-only path). Adding attention requires resolving the ggml buffer
allocation for the MTP layer, which has has_kv=false.

Approaches tried:
- build_attn with KV cache at il_kv=31: corrupts main model KV
- build_attn_inp_no_cache: GGML_ASSERT(buffer) failed
- build_attn_mha: GGML_ASSERT(buffer) failed
- Manual attention with ggml ops: GGML_ASSERT(buffer) failed

Root cause: graph scheduler doesn't allocate buffers for MTP layer
attention ops. Need to either extend n_layer_kv_from_start to include
MTP layers, or add the MTP attention to the graph plan before
scheduler runs.

Current state: FFN-only MTP gives 95% acceptance rate at temp=0.6.

Dockerfile.atlas

@@ -0,0 +1,17 @@
+FROM docker.io/nvidia/cuda:12.8.0-devel-rockylinux9 AS builder
+RUN dnf install -y cmake gcc-c++ && dnf clean all
+ENV TMPDIR=/llama.cpp/tmp
+
+# Copy local source with inline MTP changes
+COPY . /llama.cpp
+RUN cd /llama.cpp && \
+    mkdir -p /llama.cpp/tmp && \
+    cmake -B build -DGGML_CUDA=ON -DBUILD_SHARED_LIBS=OFF -DCMAKE_CUDA_ARCHITECTURES=120 -DLLAMA_BUILD_TESTS=OFF && \
+    cmake --build build --target llama-server llama-cli --config Release -j5
+
+FROM docker.io/nvidia/cuda:12.8.0-runtime-rockylinux9
+COPY --from=builder /llama.cpp/build/bin/llama-server /usr/local/bin/
+COPY --from=builder /llama.cpp/build/bin/llama-cli /usr/local/bin/
+RUN mkdir -p /models /templates
+EXPOSE 8000
+ENTRYPOINT ["/entrypoint.sh"]