Extend LLM provider coverage with two new adapters for local model
deployment:

- Ollama LLM adapter using ollixir client library with streaming support
- Ollama embedder adapter for local embedding generation
- vLLM adapter using OpenAI-compatible API via openai_ex
- Telemetry context module for lineage tracking (trace_id, work_id,
  plan_id, step_id) across all LLM adapters
- BackendBridge for CrucibleIR backend prompt/completion conversion

Dependencies updated from path-based development references to hex
package versions for portfolio_core, foundation, gemini_ex,
claude_agent_sdk, and codex_sdk. Added ollixir ~> 0.1.0 for Ollama
integration.

Includes runnable examples, setup scripts for pulling required Ollama
models, and comprehensive test coverage with Mox-based SDK mocking.