[Prefix Cache] Include lora_name in BlockStored event for deterministic KV-cache reconstruction

[sagiahrac]

Purpose

Currently, the BlockStored KV event records the lora_id associated with a cache block.
This PR replaces the use of lora_id with the lora_name in BlockStored events, aligning the event metadata with the new deterministic hashing scheme introduced in #27211.
With this change, the same block hash can be reliably reproduced across runs and instances, allowing cache lookups, routing, and sharing.

Test Plan

The relevant test files were updated.

Test Result

All updated tests pass.

[mergify]

Documentation preview: https://vllm--27577.org.readthedocs.build/en/27577/

[gemini-code-assist]

Code Review

This pull request replaces lora_id with lora_name in the BlockStored KV event to align with the deterministic hashing scheme. The code changes involve modifications in kv_events.py, kv_transfer/kv_connector/v1/offloading_connector.py, v1/core/block_pool.py, examples/online_serving/kv_events_subscriber.py, tests/v1/engine/test_engine_core_client.py, and tests/v1/kv_connector/unit/test_offloading_connector.py. The changes look correct and necessary for the intended purpose.

[vMaroon]

I think beyond reproducing KV-block hashes under the KV-cache interoperability work, the BlockStored event carries in it the vLLM hashes, and all the information used in cooking them: token-ids, block-size, and previously lora-id. After #27211, the latter changed to lora-name, therefore it must be changed here too.