Propagate chat config attributes to `chat.sendMessage()`/`chat.sendMessageStream()` spans for Google GenAI

[nicohrubec]

With the removal of the chats.create() span (which was a local object construction, not an LLM call), we lost some span attributes that were previously captured from the config parameter passed to chats.create(). These config values are reused for every subsequent chat.sendMessage() / chat.sendMessageStream() call on that chat instance and should be welded onto those spans instead.

Missing attributes on chat.sendMessage() / chat.sendMessageStream() spans:

• gen_ai.request.model — from model
• gen_ai.request.temperature — from config.temperature
• gen_ai.request.top_p — from config.topP
• gen_ai.request.top_k — from config.topK
• gen_ai.request.max_tokens — from config.maxOutputTokens
• gen_ai.request.frequency_penalty — from config.frequencyPenalty
• gen_ai.request.presence_penalty — from config.presencePenalty
• gen_ai.request.available_tools — from config.tools
• gen_ai.system_instructions — from config.systemInstruction

Solution brainstorm: The relevant parameters should be captured during the chats.create() call and merged onto the LLM invocation spans.

[linear-code]

JS-2041