feat(provider): add LocalAI and Ollama providers

[devin-ai-integration]

Description

Adds two new built-in providers for self-hosted / local AI servers:

• SmartHopper.Providers.LocalAI — talks to a LocalAI instance via its standard OpenAI-compatible POST /v1/chat/completions endpoint.
• SmartHopper.Providers.Ollama — talks to an Ollama instance via its OpenAI-compatible endpoint (POST /v1/chat/completions).

Both providers are near drop-in mirrors of the existing DeepSeek / OpenAI providers, adapted for local servers. Because LocalAI and Ollama both speak OpenAI's chat-completions wire format, almost all of the per-request encoding, decoding, tool-call handling, and SSE streaming logic is shared with the other OpenAI-compatible providers.

What each provider supports

• OpenAI-compatible messages array (system / user / assistant / tool roles, including multi-turn tool-call results)
• Tools / function calling (tools / tool_choice, parallel tool calls, tool-call results)
• JSON Schema structured outputs via response_format (json_schema) with a wrapping *JsonSchemaAdapter for non-object root schemas, mirroring the OpenAI adapter pattern
• Server-Sent Events streaming with the nested *StreamingAdapter partial class
• Optional Authorization: Bearer <token> header — useful when the local server is fronted by a reverse-proxy that requires auth, but not required for the default loopback setups

Settings (per-provider — Base URL is the key new addition)

Setting	Type	LocalAI default	Ollama default	Required
Base URL	URL	http://localhost:8080/v1	http://localhost:11434/v1	✅
API Key	string	(empty)	(empty)	❌
Model	string	(empty — user-managed)	(empty — user-managed)	❌
Enable Streaming	bool	true	true	❌
Max Tokens	int	4096	4096	❌
Temperature	double [0.0, 2.0]	0.7	0.7	❌
seed (Ollama only)	int	—	(unset)	❌

The Base URL is validated as an absolute http(s) URI; Max Tokens must be positive; Temperature is clamped to [0.0, 2.0]. Invalid values surface a styled error dialog when persisting settings, matching the pattern used by DeepSeekProviderSettings.

Model registries

Both *ProviderModels.cs return an empty list. LocalAI and Ollama users install their own models locally (e.g. ollama pull llama3.1, or any GGUF / Transformers model in a LocalAI model directory), so the provider does not attempt to enumerate them. The UI lets the user type whatever model name they have configured on the server (e.g. llama3.1, qwen2.5:14b, mistral:7b-instruct, phi3, …).

Wiring

• SmartHopper.sln — new project entries + 24 build-configuration rows
• SmartHopper.Infrastructure.csproj — two new InternalsVisibleTo entries (SmartHopper.Providers.LocalAI, SmartHopper.Providers.Ollama) with PublicKey=$(SmartHopperPublicKey), matching the existing built-in providers
• SmartHopper.Components.Test.csproj — two new ProjectReference entries so test components can resolve the providers
• yak-package/manifest.yml — adds localai / ollama keywords
• DEV.md — promotes both providers from "Planned" to "🟠 Testing" in the capabilities table and adds their *ProviderModels.cs paths to the model-registry source-file list
• README.md — adds the two providers to the supported-providers list
• CHANGELOG.md — entries under [Unreleased] → Added

Notes on tooling

• The provider model auto-update workflow (.github/workflows/chore-update-provider-models.yml) is intentionally not extended to these providers: it pulls metadata from OpenRouter as the single source of truth, and OpenRouter does not expose local LocalAI / Ollama models.
• Icon assets (localai_icon.png, ollama_icon.png) are minimal 16×16 placeholders to match the existing convention; they can be replaced with branded artwork in a follow-up.

Breaking Changes

None — both providers are purely additive.

Testing Done

• Followed the existing OpenAI / DeepSeek provider patterns exactly (same AIProvider<T> singleton + AIProviderSettings + IAIProviderFactory + *ProviderModels + *JsonSchemaAdapter + nested *StreamingAdapter partial class).
• Compiled both projects locally on Linux against net7.0 (SmartHopper.Providers.LocalAI, SmartHopper.Providers.Ollama) — both build with 0 errors after wiring up the Infrastructure InternalsVisibleTo entries. The remaining build warnings are pre-existing analyzer warnings from SmartHopper.Infrastructure and are not introduced by this change.
• End-to-end runtime verification against live LocalAI / Ollama servers is left for follow-up since it requires those servers to be installed on the developer machine.

Checklist

• This PR is focused on a single feature or bug fix
• Version in Solution.props was updated, if necessary, and follows semantic versioning (not needed — additive providers, no public-API or behavior change)
• CHANGELOG.md has been updated
• PR title follows Conventional Commits format
• PR description follows Pull Request Description Template

Link to Devin session: https://app.devin.ai/sessions/9ce31e9f6c7140d5adfcdc9c557b4a3f
Requested by: @marc-romu

[devin-ai-integration]

🤖 Devin AI Engineer

I'll be helping with this pull request! Here's what you should know:

✅ I will automatically:

• Address comments on this PR that start with 'DevinAI' or '@devin'.
• Look at CI failures and help fix them

Note: I can only respond to comments from users who have write access to this repository.

⚙️ Control Options:

• Disable automatic comment and CI monitoring