optimize Hash for Path

Hashing does not have to use the whole Components parsing machinery because we only need it to match the
normalizations that Components does.

* stripping redundant separators -> skipping separators
* stripping redundant '.' directories -> skipping '.' following after a separator

That's all it takes.

And instead of hashing individual slices for each component we feed the bytes directly into the hasher which avoids
hashing the length of each component in addition to its contents.

library/std/src/path.rs

@@ -2873,9 +2873,35 @@ impl cmp::PartialEq for Path {
 #[stable(feature = "rust1", since = "1.0.0")]
 impl Hash for Path {
     fn hash<H: Hasher>(&self, h: &mut H) {
-        for component in self.components() {
-            component.hash(h);
+        let bytes = self.as_u8_slice();
+
+        let mut component_start = 0;
+        let mut bytes_hashed = 0;
+
+        for i in 0..bytes.len() {
+            if is_sep_byte(bytes[i]) {
+                if i > component_start {
+                    let to_hash = &bytes[component_start..i];
+                    h.write(to_hash);
+                    bytes_hashed += to_hash.len();
+                }
+
+                // skip over separator and optionally a following CurDir item
+                // since components() would normalize these away
+                component_start = i + match bytes[i..] {
+                    [_, b'.', b'/', ..] | [_, b'.'] => 2,
+                    _ => 1,
+                };
+            }
+        }
+
+        if component_start < bytes.len() {
+            let to_hash = &bytes[component_start..];
+            h.write(to_hash);
+            bytes_hashed += to_hash.len();
         }
+
+        h.write_usize(bytes_hashed);
     }
 }