feat: update llama.cpp to 936918514

[abetlen]

Updates the vendored llama.cpp revision and synchronizes the affected Python bindings.

• Advances llama.cpp from e3546c794 to 936918514.
• Synchronizes load-mode, MTP loading, suppress-token, sampler, and MTMD bindings.
• Loads MTP tensors only when the example server enables draft-mtp.
• Ensures contexts close before their backing models during cyclic finalization.
• Includes upstream ggml 0.19.0 and MTMD image-scaling fixes.
• Validates the CPU and MTMD build, ctypes symbols, lint checks, and the full Python test suite.