Default to llama.cpp backend instead of platform default

[mmonteleone]

This PR changes the default behavior of corral search. Instead of defaulting to a platform-specific backend (MLX on Apple Silicon, llama.cpp elsewhere), it now consistently defaults to llama.cpp (GGUF).

• Provides more predictable search results across different platforms.
• Eliminates warnings when using the --quants flag on macOS Apple Silicon.
• README.md: Updated documentation to reflect that corral search now defaults to llama.cpp/GGUF results.
• src/lib/corral-helpers.sh: Changed platform_default_backend to return llama.cpp as the default backend globally.
• src/lib/corral-runtime.sh: Updated backend inference logic to fall back to llama.cpp by default.
• src/lib/corral-search.sh: Modified cmd_search to explicitly use llama.cpp when no --backend flag is provided.
• tests/: Updated smoke and unit tests to verify the new default search behavior and ensure no warnings are issued on Apple Silicon.

If/when MLX-LM returns from hibernation, may revisit defaulting to MLX.