cuda: infrastructure for whole-step CUDA graph capture

- Thread-local launch stream (xserv_cuda::stream): every kernel
  wrapper, cublasSetStream, and NCCL collective now launches on
  current_stream_raw() — the legacy null stream by default (behavior
  unchanged), or the capture stream installed via push_stream during
  graph capture. Capture is impossible on the legacy stream.
- Allocator retain mode: blocks freed inside a retain window are
  quarantined (RetainedBlocks) instead of pooled, so an instantiated
  graph keeps exclusive ownership of every intermediate buffer it
  references across replays.
- Capture mode GLOBAL -> THREAD_LOCAL: concurrent TP rank threads
  must not poison each other's captures with their own cudaMallocs.
- embedding_device_ids / rope_inplace_device_pos: variants reading
  token ids / positions from persistent device buffers, replacing the
  per-call host upload that a captured region cannot contain.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

crates/xserv-kernels/src/lib.rs

@@ -16,11 +16,11 @@ pub use activation::{add, bias_add_2d, gelu, gpt_oss_glu, mul, scale, silu, silu
 pub use argmax::{argmax_bf16_single, argmax_bf16_to_host};
 pub use transpose::{merge_heads_gpu, repeat_kv_gpu, reshape_heads_gpu, strided_to_contiguous_gpu, transpose_for_rope_gpu, transpose_from_rope_gpu};
 pub use attention::{attention, decode_attention, flash_attention, flash_attention_sinks, paged_decode_attention, paged_decode_attention_sinks, reshape_and_cache_bf16, reshape_and_cache_batched_bf16};
-pub use embedding::embedding;
+pub use embedding::{embedding, embedding_device_ids};
 pub use gemm::{batched_matmul, matmul, GemmBackend};
 pub use layernorm::layernorm;
 pub use rmsnorm::{add_rmsnorm, rmsnorm};
-pub use rope::{rope_inplace, RopeCache};
+pub use rope::{rope_inplace, rope_inplace_device_pos, RopeCache};
 pub use softmax::softmax;
 
 /// Register GPU kernels with the tensor crate. Call once at startup.