phase 10: GPU add/mul kernels + BF16 precision analysis

Kernel additions:
- add_f32/bf16, mul_f32/bf16 CUDA kernels (element-wise, on GPU)
- Refactored activation.rs with dispatch_unary/dispatch_binary helpers
- Qwen3 and GPT-2 now use GPU add/mul instead of CPU round-trips

GPT-2 add_bias also moved to GPU (broadcast via tile + GPU add)

BF16 precision analysis (docs/benchmarks/phase10-qwen3.md):
- Root cause: separate attention kernels materialize BF16 intermediates
  (QK^T→BF16→scale→BF16→mask→BF16→softmax→BF16 vs HF's fused FP32 path)
- HF itself SDPA vs Eager also differs by ~0.125 logit
- xserv vs HF: ~1-2 logit systematic offset, but same top-1 in 84% cases
- Industry standard for BF16: top-5 overlap (we achieve 100%)
- Fix path: Flash Attention (Phase 14) to fuse attention in FP32

Performance: TTFT 138→119ms, TBT 144→137ms (GPU ops faster than CPU)

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

crates/xserv-model/src/bin/dump-logits.rs

@@ -0,0 +1,44 @@
+use std::path::PathBuf;
+use xserv_model::{loader, KVCache, ModelConfig, Qwen3};
+use xserv_tensor::{DType, Device};
+use xserv_tokenizer::Tokenizer;
+use half::bf16;
+
+fn main() {
+    let args: Vec<String> = std::env::args().collect();
+    let model_dir = PathBuf::from(&args[1]);
+    let prompt = &args[2];
+
+    xserv_cuda::device::set_device(0).unwrap();
+    let config = ModelConfig::from_file(&model_dir.join("config.json"));
+    let weights = loader::load_model_dir(&model_dir, Device::Cuda(0));
+    let model = Qwen3::from_weights(config.clone(), weights);
+    let tokenizer = Tokenizer::from_file(&model_dir.join("tokenizer.json"));
+
+    let token_ids = tokenizer.encode(prompt);
+    eprintln!("Prompt: {prompt}");
+    eprintln!("Token IDs: {token_ids:?}");
+
+    let mut cache = KVCache::new(
+        config.num_layers(), config.num_kv_heads(), config.head_dim(),
+        DType::BF16, Device::Cuda(0),
+    );
+    let logits = model.forward_with_cache(&token_ids, &mut cache);
+    let logits_cpu = logits.to_device(Device::Cpu);
+    let data = logits_cpu.as_slice::<bf16>();
+    let vocab_size = logits.shape()[1];
+    let seq_len = logits.shape()[0];
+
+    // Print top-20 logits for the last position
+    let last_row = &data[(seq_len - 1) * vocab_size..seq_len * vocab_size];
+    let mut indexed: Vec<(usize, f32)> = last_row.iter().enumerate()
+        .map(|(i, v)| (i, v.to_f32()))
+        .collect();
+    indexed.sort_by(|a, b| b.1.partial_cmp(&a.1).unwrap());
+
+    println!("Top-20 logits (last position):");
+    for (rank, (id, val)) in indexed.iter().take(20).enumerate() {
+        let tok = tokenizer.decode(&[*id as u32]);
+        println!("  [{rank:>2}] id={id:>6} logit={val:>10.4} token={tok:?}");
+    }
+}