feat: update llama.cpp to 42e98813e

[abetlen]

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

• Advances llama.cpp from e3546c794 to 42e98813e.
• Synchronizes load-mode, suppress-token, sampler, and MTMD bindings with the current exported headers.
• Loads MTP tensors only when the example server enables draft-mtp.
• Ensures contexts close before their backing models during cyclic finalization.
• Validates the CPU and MTMD build, ctypes symbols, lint checks, and the full Python test suite.