Add ollama service - #3306

Adds services.ollama, which runs an Ollama server for local large language models. Ported from sagikazarmark/devenv-extras.

{
  services.ollama = {
    enable = true;
    loadModels = [ "qwen3:0.6b" ];
  };
}

What it does

• Runs ollama serve as a process with a port assigned by devenv (base port 11434). The process counts as ready once GET /api/version responds.
• Exports OLLAMA_HOST so the ollama CLI talks to the managed server, and OLLAMA_BASE_URL (http://…/v1) for OpenAI-compatible clients. If the server binds to 0.0.0.0 or ::, clients connect through 127.0.0.1.
• Stores models in .devenv/state/ollama/models by default. Set modelsDir to a shared directory such as ~/.ollama/models to reuse models across projects.
• If loadModels is set, a devenv:ollama:load-models task pulls those models after the server is ready and skips any that are already present. devenv up ollama runs the task too. On devenv CLIs older than 2.3.2 the module still evaluates, but models aren't pulled automatically.

Options

Changes

• src/modules/services/ollama.nix: the new module
• tests/ollama/: integration test that checks /api/version, /v1/models, ollama list and that the models directory is created
• Docs: generated options data and the service page, plus an Ollama entry with the Simple Icons logo in the technology registry

Testing

• devenv-run-tests run tests --only ollama passes on Linux.
• devenv tasks list with loadModels set and host = "0.0.0.0" shows the expected task order: setup, then the server process, then load-models. OLLAMA_HOST resolves to 127.0.0.1:<port> in that case.
• Not tested: actually pulling a model (the test leaves it out because it needs the network and is slow), macOS, and the GPU packages.