refactor(core): Introduce provider architecture

[hanXen]

Related to #306

1. What this PR does / Why we need it

This Pull Request addresses issue #306 by refactoring the core architecture of aix to support multiple LLM providers. It decouples the application from its tight integration with OpenAI by introducing a generic Provider interface.

This PR focuses exclusively on the foundational architectural changes and does not yet add a new provider. The primary goal here is to make the codebase more scalable, maintainable, and future-proof without altering any existing functionality.

A subsequent PR will leverage this new architecture to add support for local LLMs via Ollama.

2. Key Architectural Changes

To address the limitations of the original design, we've implemented a provider-centric architecture. The key changes are as follows:

2.1. Provider Interface

• We introduced a new Provider interface in internal/provider/provider.go.
• This interface defines a common contract that all LLM providers must adhere to, standardizing core operations:

  type Provider interface {
      Name() string
      ListModels(ctx context.Context) ([]string, error)
      Chat(ctx context.Context, prompt string, systemPrompt string, opts ChatOptions) (Result, error)
      ChatStream(ctx context.Context, prompt string, systemPrompt string, opts ChatOptions) (StreamResult, error)
  }

2.2. OpenAIProvider

• The original OpenAI logic from runner.go was migrated into a new internal/provider/openai.go file.
• This new OpenAIProvider fully implements the Provider interface, encapsulating all OpenAI-specific logic.

2.3. Refactored runner.go

• The Runner now acts as a high-level dispatcher.
• It is no longer concerned with the implementation details of any specific provider. Based on the --provider flag, it will instantiate the appropriate provider and delegate all LLM-related tasks to it.

2.4. Unified and Improved CLI Flags

• --provider: A new flag to select the desired provider. In this PR, it only supports the default openai.
• --model Consolidation: The --gpt3 and --gpt4 flags have been removed. Model selection is now handled exclusively by the --model flag for a consistent experience (e.g., aix --model gpt-4-turbo).

  This change is backward-compatible for typical usage. Scripts explicitly using --gpt3 or --gpt4 may need to update.

---

3. Verification: No Functionality Change

A primary goal of this refactoring was to ensure no existing functionality was lost. All features from the original OpenAI-only implementation have been successfully carried over.

Original Feature	Current Status & Analysis
API Key Handling	Preserved. The logic for handling the OpenAI API key remains unchanged.
Model Selection	Improved. The --gpt3/--gpt4 flags were deprecated in favor of a single, clearer --model flag. This is a non-breaking improvement in user experience and extensibility.
--list-models	Preserved. This flag continues to list models for OpenAI as before.
Chat (Non-Streaming)	Preserved. All parameters, including Temperature and TopP, are correctly passed to the OpenAI client via the new ChatOptions struct.
Chat (Streaming)	Preserved. The core logic using go-openai's streaming client and io.Pipe remains identical, ensuring the same performance and behavior.
System Prompt	Preserved. The --system-context (-sc) flag functions exactly as it did before.

4. Next Steps

Once this PR is approved and merged, I will submit a follow-up PR introducing the OllamaProvider. This will enable local LLM support, fully resolving issue #306.