Fix token param handling for official GPT-5 model IDs

Apply max_completion_tokens only to official GPT-5 model IDs.

Keep provider-prefixed names such as openai/gpt-5.2 on the
max_tokens path so namespaced model IDs are not treated as
official OpenAI GPT-5 identifiers.

Author: PeterDaveHello

src/services/apis/openai-token-params.mjs

@@ -1,4 +1,4 @@
-const GPT5_CHAT_COMPLETIONS_MODEL_PATTERN = /(^|\/)gpt-5([.-]|$)/
+const GPT5_CHAT_COMPLETIONS_MODEL_PATTERN = /^gpt-5([.-]|$)/
 
 function shouldUseMaxCompletionTokens(provider, model) {
   const normalizedProvider = String(provider || '').toLowerCase()

tests/unit/services/apis/openai-token-params.test.mjs

@@ -8,9 +8,9 @@ test('uses max_completion_tokens for gpt-5.x chat models', () => {
   })
 })
 
-test('uses max_completion_tokens for provider-prefixed gpt-5.x models', () => {
+test('uses max_tokens for provider-prefixed gpt-5.x model names', () => {
   assert.deepEqual(getChatCompletionsTokenParams('openai', 'openai/gpt-5.2', 2048), {
-    max_completion_tokens: 2048,
+    max_tokens: 2048,
   })
 })