perf(hash): replace twox-hash with xxhash-rust and optimize file hashing

- Replace `twox-hash` (XxHash64 via Hasher trait) with `xxhash-rust`'s
  direct `xxh64()` function call. The xxhash-rust crate has a more
  optimized implementation and the direct function avoids Hasher trait
  overhead. Same algorithm, same output, no cache invalidation.

- Pre-allocate read buffer in `git_like_hash_file` based on file
  metadata size, eliminating repeated Vec reallocations during
  `read_to_end` for every file hashed.

- Add criterion benchmarks for both turborepo-hash (task/global/file
  hash computation) and turborepo-scm (file content hashing at various
  sizes).

https://claude.ai/code/session_01JTwyUFFEGSJK1RUxGfBx8K

Author: claude

Cargo.lock

@@ -14,7 +14,7 @@ hex = "0.4.3"
 turbopath = { workspace = true }
 turborepo-lockfiles = { workspace = true }
 turborepo-types = { workspace = true }
-twox-hash = "1.6.3"
+xxhash-rust = { version = "0.8", features = ["xxh64"] }
 
 [dev-dependencies]
 test-case = { workspace = true }

crates/turborepo-hash/Cargo.toml

@@ -1,5 +1,3 @@
-use std::hash::Hasher;
-
 use capnp::message::{Allocator, Builder};
 
 pub trait Sealed<A> {}
@@ -31,9 +29,7 @@ where
 
         let buf = message.get_segments_for_output()[0];
 
-        let mut hasher = twox_hash::XxHash64::with_seed(0);
-        hasher.write(buf);
-        let out = hasher.finish();
+        let out = xxhash_rust::xxh64::xxh64(buf, 0);
 
         hex::encode(out.to_be_bytes())
     }

crates/turborepo-hash/src/traits.rs

@@ -15,7 +15,11 @@ use crate::{Error, GitHashes};
 fn git_like_hash_file(path: &AbsoluteSystemPath) -> Result<String, Error> {
     let mut hasher = Sha1::new();
     let mut f = path.open()?;
-    let mut buffer = Vec::new();
+    // Pre-allocate the buffer based on file metadata to avoid repeated
+    // reallocations during read_to_end. The +1 accounts for read_to_end's
+    // probe read that confirms EOF.
+    let estimated_size = f.metadata().map(|m| m.len() as usize + 1).unwrap_or(0);
+    let mut buffer = Vec::with_capacity(estimated_size);
     // Note that read_to_end reads the target if f is a symlink. Currently, this can
     // happen when we are hashing a specific set of files, which in turn only
     // happens for handling dotEnv files. It is likely that in the future we