[Feature] Ollama and local embedding provider support

[Siddhant-K-code]

Summary

Add support for Ollama and other local embedding models so users can run Distill without sending data to external APIs.

Motivation

• Privacy-sensitive users and enterprises can't send data to OpenAI
• Local models eliminate API costs entirely
• Ollama is the most popular local model runner (100K+ GitHub stars)
• Positions Distill as a fully self-contained, air-gapped solution

Embedding providers to add

Provider	Interface	Priority
Ollama	HTTP (/api/embeddings)	P0
HuggingFace TEI	HTTP (OpenAI-compatible)	P1
Sentence Transformers	Python subprocess	P2

Implementation

// pkg/embedding/ollama/client.go
type Config struct {
    Host  string // default: http://localhost:11434
    Model string // default: nomic-embed-text
}

type Client struct { ... }

func (c *Client) Embed(ctx context.Context, text string) ([]float32, error)
func (c *Client) EmbedBatch(ctx context.Context, texts []string) ([][]float32, error)

CLI usage

# Use Ollama for embeddings
distill api --embedding-provider ollama --embedding-model nomic-embed-text

# Use with config file
# distill.yaml
embedding:
  provider: "ollama"
  model: "nomic-embed-text"
  host: "http://localhost:11434"

Config file support

Update pkg/config/config.go to support provider field in embedding config.

Deliverables

• pkg/embedding/ollama/client.go - Ollama HTTP client
• pkg/embedding/ollama/client_test.go - Unit tests (mock HTTP)
• pkg/embedding/provider.go - Provider interface + factory function
• Update cmd/api.go and cmd/serve.go to accept --embedding-provider flag
• Update pkg/config/config.go with provider field
• Update README with Ollama setup instructions
• docker-compose.yml example with Ollama sidecar

Acceptance Criteria

• distill api --embedding-provider ollama works with local Ollama
• No data leaves the machine
• Batch embedding support
• Graceful error when Ollama is not running
• Config file supports provider selection