Fix delta role error when using custom LLM (#3223)

* Fix delta role error when using custom LLM

This addresses a delta chunk issue that happens when you use a custom baseURL on the openai chat model. Some models like llama 2 on openrouter may have a empty delta and result in a undefined error.

Example error:
```
Cannot read properties of undefined (reading 'role')",
  "error.stack": "TypeError: Cannot read properties of undefined (reading 'role')\n    at _convertDeltaToMessageChunk (/home/ubuntu/node_modules/langchain/dist/chat_models/openai.cjs:72:24)\n    at ChatOpenAI._streamResponseChunks (/home/ubuntu/node_modules/langchain/dist/chat_models/openai.cjs:409:27)\n    at process.processTicksAndRejections (node:internal/process/task_queues:95:5)\n    at async ChatOpenAI._streamIterator (/home/ubuntu/node_modules/langchain/dist/chat_models/base.cjs:77:34)\n    at async RunnableSequence._streamIterator (/home/ubuntu/node_modules/langchain/dist/schema/runnable/base.cjs:780:30)\n    at async Object.pull (/home/ubuntu/node_modules/langchain/dist/util/stream.cjs:73:41)
```

on _convertDeltaToMessageChunk, openai works fine but when you start using models from openrouter or other baseURL llms, it doesnt take in account for empty deltas and assumes theres always a value.

I was able to test with a custom baseURL and normal openai that this works with no errors on streaming after this tweak.

* add default role

* add ? on chunk text for streaming delta

* Simpler fix

* Revert

---------

Co-authored-by: jacoblee93 <jacoblee93@gmail.com>

langchain/src/chat_models/openai.ts

@@ -449,6 +449,9 @@ export class ChatOpenAI<
       }
 
       const { delta } = choice;
+      if (!delta) {
+        continue;
+      }
       const chunk = _convertDeltaToMessageChunk(delta, defaultRole);
       defaultRole = delta.role ?? defaultRole;
       const newTokenIndices = {