fix(usage): preserve nested cached_tokens in canonicalizeUsage

buildUsage() only emits cache reads under prompt_tokens_details, so the
top-level-only read dropped the count for every Responses-format provider
(codex, grok-cli, ...), persisting cached_tokens: 0 and billing cache hits
at the full input rate. Mirror the cache_creation fallback already used
just above.

open-sse/utils/usageTracking.js

@@ -190,7 +190,10 @@ export function canonicalizeUsage(usage) {
     prompt = prompt + cached + cacheCreation;
   } else {
     // OpenAI/Gemini path (or already-canonical input): prompt already includes cached_tokens.
-    cached = num(usage.cached_tokens);
+    // Mirror the cacheCreation fallback above: buildUsage() only ever emits the
+    // nested prompt_tokens_details.cached_tokens shape, so without this the
+    // cache-read count is silently dropped on every buildUsage()-derived usage.
+    cached = num(usage.cached_tokens ?? usage.prompt_tokens_details?.cached_tokens);
   }
 
   const result = {