Make Gemini Grounded model configurable (avoid hardcoded gemini-2.0-flash quota failures)

[DeadlySilent]

In test-docker evaluation, grounded search currently fails for projects where gemini-2.0-flash free-tier quota is unavailable, even when the same key works with gemini-2.5-flash.

Observed

• models.list works (key is valid)
• generateContent:
  • gemini-2.0-flash -> 429 RESOURCE_EXHAUSTED (limit: 0)
  • gemini-2.5-flash -> 200 OK
• Spectrawl grounded search still calls hardcoded gemini-2.0-flash in src/search/engines/gemini-grounded.js.

Request

Please make grounded-search model configurable and default to a non-2.0 hardcoded path, e.g.:

• prefer search.llm.model (or dedicated search.gemini-grounded.model)
• support env override (GEMINI_GROUNDED_MODEL)
• fallback default could be gemini-2.5-flash

Why

This removes false negatives where key is valid and usable, but grounded search fails due to model-specific quota policy.

Happy to submit a PR if preferred.