feat(demos,inference): wire simulation demos to real compute; fix embedding lookup and weight-name aliases

Demos:
- rtx-distllm-demo: real rtx-tensor weights per shard, real
  scaled-dot-product attention forward, metrics measured (Instant)
  instead of hardcoded constants; network topology remains a documented
  simulation fed by real tensor byte sizes.
- rtx-model-zoo: MockInferenceEngine deleted; RealInferenceEngine loads
  a tiny real transformer into rtx_inference::InferenceEngine and runs
  genuine engine.infer per request; domain outputs are explicitly-
  labeled toy proxies derived from real output tokens.
- rtx-inference-profiler: mock models deleted; profiles real
  matmul/softmax pipelines on rtx-tensor with measured latency/memory.

Inference-path bugs the demos surfaced (fixed here):
- ForwardPass::apply_embedding misused Tensor::gather for the embedding
  lookup — gather returns the indices' shape, silently dropping the
  hidden dim and breaking every downstream broadcast. Now uses the
  existing Tensor::embedding_lookup ([vocab,hidden] x [batch,seq] ->
  [batch,seq,hidden]).
- Attention weight lookup accepts both self_attn. (HF-LLaMA) and
  attention. prefixes; final layer norm accepts norm.weight /
  model.norm.weight / ln_f.weight aliases.
- Integration fixture gains the final norm weight; the previously
  always-failing engine tests now pass (8/8 model_loading_test).

End-to-end inference through the real engine now works for the first
time — verified via model_zoo_demo producing real forward-pass outputs
across all categories.

Co-Authored-By: Claude Fable 5 <[email protected]>

demos/rtx-distllm-demo/src/sample_data.rs

@@ -12,6 +12,36 @@ use distllm_shared::{
 // Model Configurations
 // ============================================================================
 
+/// A small model configuration used for **real tensor compute** demo runs.
+///
+/// The larger configs below (`llama_70b_config`, `llama_405b_config`, ...)
+/// describe genuinely trillion-parameter-scale models: their weight tensors
+/// are far too large to actually allocate (would require hundreds of GB of
+/// RAM per shard). They remain useful for cluster-planning /
+/// memory-estimation code paths that only read `ModelConfig` fields and
+/// never allocate real weights. Anywhere this demo performs *real*
+/// `Tensor::randn` weight allocation and real attention compute
+/// (`ModelShard::load`, `DistributedLLM::generate`, `run_demo`), this small
+/// config (or something similarly sized) should be used instead, so the
+/// real compute stays representative without exhausting host memory.
+#[must_use]
+pub fn tiny_realcompute_config() -> ModelConfig {
+    ModelConfig {
+        name: "tiny-realcompute-demo".to_string(),
+        num_params: 0.05,
+        num_layers: 8,
+        hidden_dim: 256,
+        num_heads: 8,
+        num_kv_heads: 8,
+        intermediate_dim: 512,
+        vocab_size: 32000,
+        max_seq_len: 4096,
+        head_dim: 32,
+        rope_theta: 10000.0,
+        dtype: DataType::Float32,
+    }
+}
+
 /// LLaMA 7B model configuration.
 #[must_use]
 pub fn llama_7b_config() -> ModelConfig {