[Core] Use sha256 bytes instead of BlockHash to reduce GC overhead (#23673)

Signed-off-by: linzebing <linzebing1995@gmail.com>

vllm/config/cache.py

@@ -24,7 +24,7 @@ logger = init_logger(__name__)
 BlockSize = Literal[1, 8, 16, 32, 64, 128]
 CacheDType = Literal["auto", "fp8", "fp8_e4m3", "fp8_e5m2", "fp8_inc"]
 MambaDType = Literal["auto", "float32"]
-PrefixCachingHashAlgo = Literal["builtin", "sha256", "sha256_cbor_64bit"]
+PrefixCachingHashAlgo = Literal["sha256", "sha256_cbor"]
 
 
 @config
@@ -63,17 +63,12 @@ class CacheConfig:
     """Sliding window size for the KV cache. This is primarily set in
     `ModelConfig` and that value should be manually duplicated here."""
     enable_prefix_caching: Optional[bool] = None
-    """Whether to enable prefix caching. Disabled by default for V0. Enabled by
-    default for V1."""
-    prefix_caching_hash_algo: PrefixCachingHashAlgo = "builtin"
+    """Whether to enable prefix caching. Enabled by default for V1."""
+    prefix_caching_hash_algo: PrefixCachingHashAlgo = "sha256"
     """Set the hash algorithm for prefix caching:\n
-    - "builtin" is Python's built-in hash.\n
-    - "sha256" is collision resistant but with certain overheads.
-    This option uses Pickle for object serialization before hashing.\n
-    - "sha256_cbor_64bit" provides a reproducible, cross-language compatible
-    hash. It serializes objects using canonical CBOR and hashes them with
-    SHA-256. The resulting hash consists of the lower 64 bits of the SHA-256
-    digest."""
+    - "sha256" uses Pickle for object serialization before hashing.\n
+    - "sha256_cbor" provides a reproducible, cross-language compatible hash. It
+    serializes objects using canonical CBOR and hashes them with SHA-256."""
     cpu_offload_gb: float = 0
     """The space in GiB to offload to CPU, per GPU. Default is 0, which means
     no offloading. Intuitively, this argument can be seen as a virtual way to