Validate GGUF tensor dimensions

[abhirup7]

Fixes #3358.

Summary

• Reject GGUF tensors whose ndim exceeds GGUF_TENSOR_MAX_DIM before converting tensor dimensions to an MLX shape, including both values in the error message.
• Add a regression test for malformed GGUF tensor rank.
• Compile gguflib sources directly into mlx via target_sources so downstream static-link consumers don't hit unresolved symbols, and expose gguflib headers only to MLX sources and the GGUF-specific test via BUILD_INTERFACE.

Tests

clang-format -i mlx/io/gguf.cpp tests/load_tests.cpp
cmake -S . -B build-cpu -DMLX_BUILD_TESTS=ON -DMLX_BUILD_PYTHON_BINDINGS=OFF -DMLX_BUILD_METAL=OFF -DMLX_BUILD_EXAMPLES=OFF -DMLX_BUILD_BENCHMARKS=OFF
cmake --build build-cpu --target tests -j8
./build-cpu/tests/tests -tc="test gguf tensor ndim validation"
./build-cpu/tests/tests -tc="test gguf*"
git diff --check