fix(search): make Gemini grounded model configurable (avoid hardcoded 2.0)

[DeadlySilent]

Summary

Fix grounded search model selection so it no longer hardcodes gemini-2.0-flash.

Problem

In some free-tier projects, gemini-2.0-flash is quota-blocked (limit: 0) while other Gemini models (e.g. gemini-2.5-flash) are available. Spectrawl grounded search still fails because src/search/engines/gemini-grounded.js hardcoded 2.0.

Changes

• src/search/index.js
  • add this.groundedModel resolution with precedence:
    1. search.llm.model
    2. search["gemini-grounded"].model
    3. search.gemini.model
    4. GEMINI_GROUNDED_MODEL
    5. GEMINI_MODEL
    6. fallback gemini-2.5-flash
  • add _engineConfig(engineName) so gemini engines receive merged config (model, apiKey).
• src/search/engines/gemini-grounded.js
  • remove hardcoded gemini-2.0-flash fallback and use configurable model/env fallback.

Validation

In isolated docker trial:

• Before patch: grounded search returned 429 quota errors on gemini-2.0-flash; no sources.
• After patch with same key/config (gemini-2.5-flash):
  • search returned sources (5+)
  • summarize flow returned answer + sources.

Related issue

Closes #7