feat: add device config (cpu/cuda) for all enrichment models

[tmaiaroto]

Summary

Adds configurable device selection (cpu/cuda) for all enrichment models, allowing users to offload inference to GPU if available.

Changes

• schemas.py: Added field to , , and (default: "cpu")
• enrichment.py:
  • Updated EnrichmentConfig.from_dict() to pass through device configs
  • Updated EnrichmentService to use device config when loading models
  • Added validation to reject invalid device values (falls back to cpu)
• reveriecore.yaml.example: Documented the new device option for each model

Config Example

enrichment:
  classifier:
    model: "MoritzLaurer/mDeBERTa-v3-base-mnli-xnli"
    device: "cuda"  # or "cpu"
  
  embedding:
    model: "BAAI/bge-small-en-v1.5"
    device: "cuda"  # or "cpu"
  
  summarization:
    model: "sshleifer/distilbart-cnn-12-6"
    device: "cuda"  # or "cpu"

Notes

• Defaults to "cpu" for backward compatibility
• Only "cpu" and "cuda" are valid values (invalid values fall back to cpu with a warning)
• The query rewriter (GGUF model) already had a device config option in retrieval.rewriter