feat: update llama.cpp to 11924d4c1

[abetlen]

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

• Advances llama.cpp from e3546c794 to 11924d4c1.
• Preserves the existing high-level mmap and mlock behavior with the expanded load-mode API.
• Loads MTP tensors only when the example server enables draft-mtp.
• Synchronizes suppress-token and MTMD text-length bindings.
• Validates the CPU and MTMD build, ctypes ABI layouts, lint checks, the full Python suite, and an MTP chat completion.