.Net: New Feature: Support HttpClient injection for Ollama connectors (embedding and chat)

[ramonduraes]

Hello team,

We're currently using Microsoft.SemanticKernel.Connectors.Ollama (v1.54.0-alpha) to integrate local LLMs via Ollama. While the experience is great overall, we noticed that the current extension methods:

• AddOllamaEmbeddingGenerator(...)
• AddOllamaChatCompletion(...)

do not allow passing a custom HttpClient, which makes it impossible to configure critical options such as Timeout, custom headers, retry policies, or advanced diagnostics.

This is a limitation especially for local LLMs like llama3.2:3b or mistral, which may take longer than 100 seconds on first generation, leading to TaskCanceledException.

We kindly suggest either:

1. Adding overloads that support injecting a preconfigured HttpClient
2. Or allowing DI via options/configuration pattern

This change would align with .NET best practices and enable full control over networking behavior when using Ollama.

Thank you