Add new multimodal LLM serving papers

[AmberLJC]

This PR adds 2 new papers to the Multi-Modal Serving Systems section:

• HydraInfer: Hybrid Disaggregated Scheduling for Multimodal Large Language Model Serving
• Enabling Disaggregated Multi-Stage MLLM Inference via GPU-Internal Scheduling and Resource Sharing

Closes #46

Generated with Claude Code

---

Note

Updates the README.md paper list.

• Adds HydraInfer: Hybrid Disaggregated Scheduling for Multimodal Large Language Model Serving to the Multi-Modal Serving Systems section
• Adds Enabling Disaggregated Multi-Stage MLLM Inference via GPU-Internal Scheduling and Resource Sharing to the Multi-Modal Serving Systems section

^{Written by Cursor Bugbot for commit 473028e. This will update automatically on new commits. Configure here.}

[claude]

PR Review: Add new multimodal LLM serving papers

Thank you for the contribution! I've reviewed the changes and have the following feedback:

✅ Strengths

1. Correct Section Placement: Both papers are appropriately placed in the Multi-Modal Serving Systems section, which aligns with their focus on multimodal LLM inference.

2. Accurate arXiv Links: The arXiv URLs are correctly formatted without the www. prefix, following the repository's formatting guidelines.

3. Relevant Content: Both papers are highly relevant to the repository's focus on LLM systems research, specifically addressing disaggregated scheduling and GPU resource management for multimodal models.

⚠️ Issues Found

Line 319 (Second paper): The entry is missing the metadata format used throughout the repository.

Current format:

- [Enabling Disaggregated Multi-Stage MLLM Inference via GPU-Internal Scheduling and Resource Sharing](https://arxiv.org/abs/2512.17574)

Expected format (according to CLAUDE.md):

- [Paper Title](URL): Brief description | Venue/Organization

📝 Recommendations

1. Add description separator: The second paper entry should include a colon (:) followed by a brief description after the link.

2. Add metadata: If available, include the venue or organization information after a pipe (|) separator.

Suggested correction for line 319:

- [Enabling Disaggregated Multi-Stage MLLM Inference via GPU-Internal Scheduling and Resource Sharing](https://arxiv.org/abs/2512.17574): Disaggregated multi-stage MLLM inference with GPU-internal scheduling

📋 Minor Observations

• Line 296: I noticed an existing inconsistency in the file - Optimus: has a colon in the title itself, which is unusual. This is not related to your PR but something to note for future consistency.

• Line 307: Similarly, DistriFusion: has this same pattern. Your entries actually follow better practices than some existing entries.

✅ Overall Assessment

Code Quality: Good - follows most formatting conventions
Best Practices: Mostly adhered to, with one formatting issue
Security: N/A - documentation only
Test Coverage: N/A - documentation only
Performance: N/A - documentation only

Recommendation: Merge with minor revision to add the description separator and metadata to the second paper entry for consistency with repository guidelines.

---

Review generated following the guidelines in CLAUDE.md