Add bounded graph primitives RFC

[jasonmassie01]

Summary

This PR adds a proposed RFC for a bounded graph primitives module for Valkey.

The core framing is:

• module-only, not Valkey core;
• no Cypher, Gremlin, SPARQL, query planner, arbitrary pattern matching, shortest path, or global graph algorithms;
• native Valkey Sets/Hashes first, no custom graph type in v0.1;
• a real graph key as the first command argument and cluster-slot anchor;
• directed edges by default;
• bounded batch writes, neighbor reads, BFS expansion, edge scan, graph drop, and invariant checks;
• metadata-first replies that report truncation, work performed, and elapsed time.

The product line I am trying to hold is:

Make small graph neighborhoods safe to expand and safe to operate, without turning Valkey into a graph database.

Files

• GRAPH_PRIMITIVES.md: RFC proposal in the Valkey RFC template.
• GRAPH_PRIMITIVES_PRFAQ.md: supporting PRFAQ with customer framing, critic objections, managed-cloud adoption concerns, and launch metrics.

Maintainer Feedback Requested

1. Does this bounded-primitives shape align with Valkey module direction?
2. Is the richer v0.1 lifecycle justified, or should scan/drop/check be deferred?
3. Are source/destination catalogs worth the write amplification?
4. Should traversal metadata be default-on with NOMETA as the opt-out?
5. Are explicit over-ceiling errors preferable to silent clamping?
6. Are the proposed command flags, key specs, and ACL categories appropriate?
7. What additional constraints would make this more suitable for provider-bundled managed Valkey services?

Validation

• Checked the RFC against the repository template.
• Checked for placeholders and non-ASCII artifacts.
• Confirmed both files are Markdown-only proposal material.