Bump ai-sdk deps and update AI configs

- Bumped @ai-sdk packages and prettier; added
  ai-sdk-provider-opencode-sdk
- Export new convexMemory using ModelRouterEmbeddingModel and
  semantic/workingMemory options
- Change Lance/Mongo defaults to 'mastra_vectors' and set Lance
  embedding
  dimension to 3072; simplify Lance working memory template
- Update Gemini CLI: use thinkingConfig, add verbose/logger options
- Switch Google model calls from languageModel to chat (incl. image
  model)
- Add ResearchPhase type and make researchPhase field type-safe
- Add typed providerOptions for Google/OpenAI in mastra initialization