fix(api-nodes): edge cases in responses for Gemini models

[bigcat88]

1. Gemini-3-Pro-Image: When the user enters a help-style prompt (e.g., help, help me) with response modality set to IMAGE, the API may return promptFeedback=None. This fix detects that case and surfaces the appropriate error message instead of failing silently.

2. Gemini-2.5-flash-image: With modality=IMAGE and a help-style prompt, the response can contain usageMetadata.candidatesTokensDetails=None. Add a null guard so we don’t attempt to enumerate a NoneType and spam the logs.

3. Docs: Update the aspect_ratio description for Gemini-3-Pro to be more accurate.

4. Sampling defaults: Stop hard-coding model parameters (e.g., temperature, topK) in the Pydantic model. Recent Google docs recommend leaving temperature unset and note that defaults vary by model. We were also sending topK=40 while the Gemini-3-Pro-Image model default is 64; rely on backend defaults instead. Reference: https://docs.cloud.google.com/vertex-ai/generative-ai/docs/models/gemini/3-pro-image

API Node PR Checklist

Scope

• Is API Node Change

Pricing & Billing

• Need pricing update
• No pricing update

If Need pricing update:

• Metronome rate cards updated
• Auto‑billing tests updated and passing

QA

• QA done
• QA not required

Comms

• Informed Kosinkadink