turbo: One canonical instance per type (3% faster)

src/Turbo/TurboExtensionEnabler.php

@@ -22,7 +22,7 @@
 	 * version is the short SHA of the last commit touching turbo-ext/src/,
 	 * enforced by the phar.yml turbo-version job.
 	 */
-	public const EXPECTED_EXTENSION_VERSION = 'dabbfe7';
+	public const EXPECTED_EXTENSION_VERSION = '873ede9';
 
 	private static bool $typeCombinatorCacheEnabled = false;
 

src/Type/TypeCombinatorCache.php

@@ -10,9 +10,12 @@
  * This PHP implementation performs no caching at all — it delegates straight to
  * TypeCombinator. The native implementation memoizes each operation on a
  * structural key of its arguments; roughly 91% of the calls in a
- * self-analysis run repeat an argument tuple that was already computed.
+ * self-analysis run repeat an argument tuple that was already computed. It also
+ * interns the results, keeping one canonical instance per distinct type value,
+ * so operations reaching the same value by different routes return the same
+ * object and identity checks can stand in for structural comparison.
  *
- * The cache is scoped to a single container: memoization hands back shared Type
+ * The cache is scoped to a single container: it hands back shared Type
  * instances, and Type objects carry a lazily resolved ClassReflection tied to the
  * container that created them.
  *