refactor(open-sse): extract buildUsage helper, dedup token-details (B2)

Add helpers/usageHelper.js for conditional prompt/completion token details.
Apply to gemini/codex/claude response translators; keep each provider's token
math intact. No behavior change; golden + gate: no regression.

Co-authored-by: Cursor <cursoragent@cursor.com>

open-sse/translator/helpers/usageHelper.js

@@ -0,0 +1,14 @@
+// Build OpenAI usage object. Caller computes prompt/completion/total (provider math).
+// Optional details added only when > 0 (matches existing claude/gemini/codex behavior).
+export function buildUsage({ promptTokens, completionTokens, totalTokens, cachedTokens = 0, cacheCreationTokens = 0, reasoningTokens = 0 }) {
+  const usage = { prompt_tokens: promptTokens, completion_tokens: completionTokens, total_tokens: totalTokens };
+  if (cachedTokens > 0 || cacheCreationTokens > 0) {
+    usage.prompt_tokens_details = {};
+    if (cachedTokens > 0) usage.prompt_tokens_details.cached_tokens = cachedTokens;
+    if (cacheCreationTokens > 0) usage.prompt_tokens_details.cache_creation_tokens = cacheCreationTokens;
+  }
+  if (reasoningTokens > 0) {
+    usage.completion_tokens_details = { reasoning_tokens: reasoningTokens };
+  }
+  return usage;
+}