sycl: Improve mul_mat_id memory efficiency and add BF16 fast path (#22119)

* sycl: size mul_mat_id staging buffers by routed rows

Previously src1_contiguous/dst_contiguous in ggml_sycl_mul_mat_id were
sized to ggml_nelements(src1/dst), which over-allocates when ne12 > 1
and can fail with UR_RESULT_ERROR_OUT_OF_HOST_MEMORY on Level Zero for
MoE models (notably with --cpu-moe). Size them by the actual number of
routed rows (ids->ne[1] * n_ids) instead.

* sycl: add bf16 mul_mat fast path via DNNL

When src0 is BF16 (commonly the case for lm_head / output.weight), the
existing f16 path is skipped because bf16 isn't covered, and the f32
fallback dequantizes the entire src0 slab to f32 in a single pool alloc
(row_diff*ne00 floats). For large-vocab models this can reach several
GB and fail with UR_RESULT_ERROR_OUT_OF_HOST_MEMORY on Level Zero.

Add a bf16xbf16 -> f32 DNNL matmul fast path that uses the bf16 storage
in place and only materializes a small src1 bf16 conversion buffer. bf16
matmul accumulates in f32, so it's correct even when the op requests
GGML_PREC_F32 (as lm_head does).

- gemm.hpp: map bfloat16 to dnnl::memory::data_type::bf16.
- convert.{hpp,cpp}: expose ggml_get_to_bf16_sycl for f32/f16/bf16 -> bf16.
- ggml-sycl.cpp: take the bf16 path early in ggml_sycl_op_mul_mat_sycl
  when DNNL and GGML_SYCL_HAS_BF16 are both available.

ggml/src/ggml-sycl/convert.hpp

@@ -23,6 +23,11 @@ typedef to_t_sycl_t<sycl::half> to_fp16_sycl_t;
 to_fp16_sycl_t ggml_get_to_fp16_sycl(ggml_type type, ggml_tensor * dst);
 to_fp32_sycl_t ggml_get_to_fp32_sycl(ggml_type type, ggml_tensor * dst);
 
+#ifdef GGML_SYCL_HAS_BF16
+typedef to_t_sycl_t<sycl::ext::oneapi::bfloat16> to_bf16_sycl_t;
+to_bf16_sycl_t ggml_get_to_bf16_sycl(ggml_type type, ggml_tensor * dst);
+#endif
+
 // Nc = Non-contiguous
 template <typename T>
 using to_t_nc_sycl_t = void (*)(const void * x, T * y, int64_t ne00, int64_t ne01, int64_t ne02, int64_t ne03,
@@ -35,15 +40,19 @@ template<typename dst_t, typename src_t>
  inline dst_t ggml_sycl_cast(src_t x) {
     if constexpr (std::is_same_v<dst_t, src_t>) {
         return x;
+#ifdef GGML_SYCL_HAS_BF16
     } else if constexpr (std::is_same_v<dst_t, sycl::ext::oneapi::bfloat16>) {
         return sycl::ext::oneapi::bfloat16(float(x));
     } else if constexpr (std::is_same_v<src_t, sycl::ext::oneapi::bfloat16>) {
         return static_cast<float>(x);
+#endif
     } else if constexpr (std::is_same_v<src_t, sycl::float2> && std::is_same_v<dst_t, sycl::half2>) {
         return x.template convert<sycl::half, sycl::rounding_mode::rte>();
+#ifdef GGML_SYCL_HAS_BF16
     } else if constexpr (std::is_same_v<src_t, sycl::float2> &&
                          std::is_same_v<dst_t, sycl::vec<sycl::ext::oneapi::bfloat16, 2>>) {
         return {x.x, x.y};
+#endif
     } else if constexpr(std::is_same_v<dst_t, int32_t>) {
         return int32_t(x);
     } else {