Add RPG Semantic-ID recommender

[hieuddo]

Description

This PR integrates RPG (Generating Long Semantic IDs in Parallel for
Recommendation) into cornac. RPG scales up to 16-64 unordered codes for Semantic IDs (TIGER uses 4 hierarchical codes).

Key changes

• Added RPG through the NextItemRecommender interface, with graph-guided
  and exact full-catalog scoring.
• Added the official defaults and tuned configurations for Amazon Beauty,
  Sports, and Toys.
• Added aligned item-feature support to NextItemEvaluation, documentation,
  requirements, an example, and evaluation tests.

Verification and tuning

Dataset	LR	Temp.	Codebooks	Beams	Edges / steps	R@5 ours / paper	N@5 ours / paper	R@10 ours / paper	N@10 ours / paper
Beauty	0.010	0.03	32	20	200 / 3	0.0533 / 0.0550	0.0372 / 0.0381	0.0789 / 0.0809	0.0454 / 0.0464
Sports	0.003	0.03	16	100	30 / 5	0.0288 / 0.0314	0.0198 / 0.0216	0.0420 / 0.0463	0.0241 / 0.0263
Toys	0.003	0.03	16	200	20 / 3	0.0613 / 0.0592	0.0415 / 0.0401	0.0898 / 0.0869	0.0506 / 0.0490

Setting: Amazon Reviews 2014 5-core, chronological leave-last-out,
mode="last", max_len=50, seed=123, and the official RPG
text-embedding-3-large item embeddings.

The results are reasonably comparable with the paper: Beauty and Sports are
within 3.1% and 8.3% of the reported R@5, while Toys is 3.5% higher.

Checklist:

• I have added tests.
• I have updated the documentation accordingly.
• I have updated README.md (if you are adding a new model).
• I have updated examples/README.md (if you are adding a new example).