feat: add support for LoRA adapters

[oglego]

Summary

This PR adds support for LoRA adapters and closes #18. If there is anything that I should modify or change on this just let me know!

Changes

The main items that I have added on this PR are:

• LoraAdapterConfig (path, scale) and ModelConfig::loras (std::vector<LoraAdapterConfig>) for configuring one or more adapters.
• Model::initialize_context() loads and applies LoRA adapters to the model context with individual scaling via llama_set_adapters_lora().
• Model::get_loras() getter for inspecting loaded adapter pointers.
• examples/lora/ — Example application with documentation for converting adapters and using task prompt templates (including a worked example using IBM Granite RAG query rewriting).
• tests/test_lora.cpp — unit tests covering config defaults, default scale, and multiple adapter stacking.
• Root and examples README.md updates documenting LoRA usage and multi-agent context sharing.

For testing this locally and interacting with it I used the recommended model (granite-4.0-micro-Q8_0.gguf) along with a converted Granite query-rewriter LoRA adapter (query_rewrite_lora.gguf).

Testing

• ctest — all 5 suites pass (ToolTests, CallbacksTests, GrammarTests, LoraTests, ChatParserTests)
• clang-format passes
• Manual verification against a real adapter: granite-4.0-micro-Q8_0.gguf + IBM Granite RAG's query-rewrite LoRA, confirmed the adapter correctly rewrites contextual follow-ups (e.g., "What about Germany?" → "What is the capital of Germany?") that the base model alone does not

Thanks in advance for the review!!

[daavoo]

LGTM, thanks! I pushed a small README note (7a68025) explaining that prompt cache files don't record which LoRA adapters were active. Storing an adapter fingerprint with the cache can be a follow-up.

[oglego]

Thank you!!