fix(pai-art-skill): Apply model-specific default sizes in Generate.ts

[b1ackmartian]

Summary

• Fixed broken size validation logic in Generate.ts that caused errors when using models with different size formats
• Applied sensible model-specific defaults when --size is omitted or invalid

Fixes #380

Problem

The validation logic was structured as:

if (gpt-image-1 && NOT valid) throw
else if (nano-banana-pro) { validate }
else if (NOT valid for replicate) throw

When using gpt-image-1 without --size, the default "2K" would pass the first condition (since the size IS technically invalid for OpenAI), but then if you specify a valid OpenAI size like 1024x1024, the first condition becomes false and it falls through to the replicate check which fails.

Solution

Restructured to properly branch per model type and apply defaults:

• gpt-image-1: defaults to 1024x1024
• nano-banana-pro: defaults to 2K
• flux/nano-banana: defaults to 16:9

Test Plan

• Tested --model gpt-image-1 without --size - now defaults to 1024x1024
• Tested --model nano-banana-pro without --size - works with 2K default
• Generated test image successfully with Google API

---

🤖 Generated with Claude Code

[kaimagnus]

Thank you for this fix, @b1ackmartian! You nailed the issue — each model needs its own default size format.

We've applied this fix in commit 0fc1931, which adds model-specific default sizing (flux/nano-banana get aspect ratios, gpt-image-1 gets pixel dimensions, nano-banana-pro gets resolution tiers). Closing in favor of that commit. Great analysis!