`HuggingFaceRequestManager` has Llama-specific prompt instuction

**Describe the bug**
The method `HuggingFaceRequestManager.createInstructionPrompt` [here](https://github.com/JetBrains-Research/TestSpark/blob/a5c0e7a6ce0daffb01fbf522109cc78706aad613/src/main/kotlin/org/jetbrains/research/testspark/tools/llm/generation/hf/HuggingFaceRequestManager.kt#L97C17-L97C40) is Llama specific. The implementation should be rewritten so that Llama-specific features are applied only to Llama models.

**Screenshots**

Start in src/main/kotlin/org/jetbrains/research/testspark/tools/llm/generation/hf/HuggingFaceRequestManager.kt and inspect createInstructionPrompt at the referenced location. Trace how the selected model is identified and verify the current prompt behavior for Llama and non-Llama models. Done means Llama-specific features are applied only to Llama models.

Written by the indexing model from the issue text.