[Enhancement] Upgrade cache-dit from 1.2.0 to 1.3.0

[SamitHuang]

Purpose

Upgrade the cache-dit dependency from 1.2.0 to 1.3.0 (latest release). This is a version bump with full API backward compatibility — all existing imports, enable_cache(), refresh_context(), BlockAdapter, DBCacheConfig, etc., remain unchanged.

Test Plan

• Verified all cache_dit imports used in vllm_omni/diffusion/cache/cache_dit_backend.py pass with 1.3.0
• Ran offline inference benchmark on Qwen/Qwen-Image (text-to-image, 1024x1024, 50 steps) comparing with and without cache-dit acceleration
• Pre-commit passes on the changed file

# Without cache-dit (baseline)
CUDA_VISIBLE_DEVICES=1 python examples/offline_inference/text_to_image/text_to_image.py \
  --model Qwen/Qwen-Image --prompt "a cup of coffee on the table" \
  --seed 142 --num-inference-steps 50 --height 1024 --width 1024

# With cache-dit 1.3.0
CUDA_VISIBLE_DEVICES=1 python examples/offline_inference/text_to_image/text_to_image.py \
  --model Qwen/Qwen-Image --prompt "a cup of coffee on the table" \
  --seed 142 --num-inference-steps 50 --height 1024 --width 1024 \
  --cache-backend cache_dit --enable-cache-dit-summary

Test Result

Benchmark on single NVIDIA H800 GPU:

Metric	Without Cache-DiT	With Cache-DiT 1.3.0	Speedup
Total generation time	7.551s	3.761s	2.01x
Diffusion engine exec time	7,436ms	3,644ms	2.04x

Cache-dit 1.3.0 delivers ~2x acceleration on Qwen-Image with default DBCache config (Fn=1, Bn=0, W=4, threshold=0.24), consistent with 1.2.0 behavior.

---

Essential Elements of an Effective PR Description Checklist

• The purpose of the PR, such as "Fix some issue (link existing issues this PR will resolve)".
• The test plan. Please provide the test scripts & test commands.
• The test results. Please paste the results comparison before and after, or the e2e results.
• (Optional) The necessary documentation update — N/A, no doc changes needed.
• (Optional) Release notes update — N/A, minor dependency bump.