fix(exact-prefix): adapt LingBot binding to TeleFuser int KV indices

[yJader]

Background

TeleFuser PR Tele-AI/TeleFuser#3 updates the LingBot-World-Fast runtime to store KV cache progress metadata (global_end_index and local_end_index) as Python int values instead of CUDA scalar tensors. It also exposes local attention settings through the pipeline config. CacheSeek's exact-prefix LingBot binding needs to match those runtime contracts so reuse bookkeeping can read/write the same metadata shape as current TeleFuser.

Summary

• Adapt the exact-prefix LingBot binding to TeleFuser's integer global_end_index and local_end_index KV metadata.
• Update the LingBot binding tests to model direct integer KV indices.
• Select full-KV CacheSeek config in the LingBot e2e example when TeleFuser runs with local_attn_size == -1, while keeping rolling config for rolling-window runs.

Test Plan

• python -m pytest -q tests/test_world_kv_telefuser_binding.py -> 3 passed
• python examples/exact_prefix_reuse/e2e_telefuser_lingbot.py ... -> all_pass: true