Switch hash tables to emhash8::HashMap.

This is much faster than std::unordered_map, and also slightly faster
than phmap::flat_hash_map that was included in PR #2468.
It is MIT-licensed, and we just include the .h file wholesale.

I haven't done a detailed test of all the various unordered_maps
out there, but this is the overall highest-ranking contender on

  https://martin.ankerl.com/2022/08/27/hashmap-bench-01/

except for ankerl::unordered_dense::map, which requires C++17.

For a no-op build of Chromium (Linux, Zen 2),
this reduces time spent from 5.14 to 4.62 seconds.

src/hash_map.h

@@ -19,6 +19,7 @@
 #include <string.h>
 #include "string_piece.h"
 #include "util.h"
+#include "third_party/emhash/hash_table8.hpp"
 
 // MurmurHash2, by Austin Appleby
 static inline
@@ -53,8 +54,6 @@ unsigned int MurmurHash2(const void* key, size_t len) {
   return h;
 }
 
-#include <unordered_map>
-
 namespace std {
 template<>
 struct hash<StringPiece> {
@@ -73,7 +72,7 @@ struct hash<StringPiece> {
 /// mapping StringPiece => Foo*.
 template<typename V>
 struct ExternalStringHashMap {
-  typedef std::unordered_map<StringPiece, V> Type;
+  typedef emhash8::HashMap<StringPiece, V> Type;
 };
 
 #endif // NINJA_MAP_H_

src/third_party/emhash/README.ninja

@@ -0,0 +1,7 @@
+Description: emhash8::HashMap for C++14/17
+Version: 1.6.5 (commit bdebddbdce1b473bbc189178fd523ef4a876ea01)
+URL: https://github.com/ktprime/emhash
+Copyright: Copyright (c) 2021-2024 Huang Yuanbing & bailuzhou AT 163.com
+SPDX-License-Identifier: MIT
+Local changes:
+ - None.