feat: sync llama.cpp to b9297

[github-actions]

🤖 Automated llama.cpp sync

This PR was automatically created/updated by the daily sync workflow.

Changes:

• Updated llama.cpp submodule from b9254 to b9297
• Regenerated bindings and build files

Verification:

• ✅ Bootstrap script completed successfully (including iOS Metal compilation)
• ✅ iOS frameworks build completed successfully (macOS job)
• ✅ Android libraries build completed successfully (Ubuntu job)
• ✅ C++ unit tests passed
• ✅ TypeScript build completed successfully

📋 llama.cpp changes (b9254 → b9297)

• b0df4c0 model : add NVFP4 MTP scale tensors (#23563)
• a497476 ggml : Check the right iface method before using the fallback 2d get (#23514)
• 95405ac vulkan: fix windows find_package of SPIRV-Headers (#23215)
• 0f3cb3f opencl: generalize Adreno MoE kernels on M (#23449)
• 1acee6b server: only parse empty msg if continuing an assistant msg (#23506)
• ef570f6 perplexity : fix integer overflow (#23496)
• cc9e331 SYCL: improve MoE prefill throughput (#23142)
• bcfd198 sycl : Level Zero detection in ggml_sycl_init (#23097)
• 56f16f2 SYCL : gated_delta_net K>1 (#23174)
• 8cc67ef SYCL: add BF16 to DMMV kernel path (~4x tg speedup on Intel Arc) (#21580)
• 95feeab docs: Update documentation with Granite 4.0/4.1 (#23404)
• 99d4026 ggml-zendnn : add Q8_0 quantization support (#23414)
• 9c92e96 cmake : build router app only during standalone builds (#23521)
• afcda09 vocab : fix HybridDNA tokenizer (#23466)
• bbce619 cmake : add install() for impl libraries + fix apple builds (#23511)
• 4f0e43d CUDA: fix PDL CC check for JIT compilation (#23471)
• bb28c1f cmake : remove STATIC from impl libraries, enable LLAMA_BUILD_APP by default (#23462)
• ee7c305 Update WebGPU support and add link to blog/demo (#23483)
• 47c0eda vulkan: fuse snake activation (mul, sin, sqr, mul, add) (#22855)
• 5306f4b fix(flash-attn): replace f32 with kv_type and q_type (#23372)
• 40d5358 tests : move save-load-state from examples to tests (#23336)
• b65bb4b server: expose prompt token counts in /slots endpoint (#23454)
• a1a69f7 metal : optimize concat kernel and fix set kernel threads (#23411)
• 52fb93a server : free draft/MTP resources on sleep to fix VRAM leak (#23461)
• c902171 server: re-inject subcommand when router spawns children under unified binary (#23442)
• 1d7ab2b app : add batched-bench, fit-params, quantize & perplexity (#23459)
• 12e5d99 mtp: use inp_out_ids for skipping logit computation (#23433)
• 7ea23dd vocab : add Carbon-3B (HybridDNATokenizer) support (#23410)
• 2fc8d18 doc: fix spec mtp typo (#23435)
• 5e932a1 ui: Improve Git Hooks for UI development (#23403)
• 2754ce1 ggml : Check the right iface method before using the fallback 2d get (#23306)
• eeeaf61 llama-graph: fix null-buffer crash in llm_graph_input_attn_kv_iswa for SWA-only models (#23131)
• 0be8468 hexagon: ssm-conv fix for large prompts (#23307)
• ce02093 app : show version (#23426)
• 6a257d4 mtmd, model : merge HunyuanOCR into HunyuanVL and fix OCR vision precision (#23329)
• 3a479c9 ui: Add max image size option (#22849)
• ad27757 Move to backend sampling for MTP draft path (#23287)
• 3a6db74 opencl: refactor backend initilization (#23318)
• 510b5c2 common/speculative : fix nullptr crash in get_devices_str (#23386)
• a8681a0 mtmd : DeepSeek-OCR image processing fixes, img_tool::resize padding refactor (#23345)
• acd604f vulkan: optimize operations in the IM2COL shader (#22685)
• 6ce9671 feat: Add WAV MIME type variants and improve audio format detection (#23396)
• c9872a2 hexagon: HMX quantized matmul rework (#23368)

Please review and merge if all checks pass.