fix(session): gate screenshot media for visionless models

[Alezander9]

Summary:

• remove unsupported tool-result media from model messages based on selected model input capabilities
• replace browser screenshots for visionless models with a short text notice
• keep media attachments for vision-capable models

Tests:

• bun test test/session/message-v2.test.ts
• bun typecheck
• pre-push hook: bun turbo typecheck filtered packages

---

Summary by cubic

Gates tool-result media by model input capabilities. Visionless models no longer receive screenshots/images; they get a short text notice, while vision-capable models keep media.

• Bug Fixes
  • Filter attachments by model.capabilities.input (image/audio/video/pdf); add a single, deduped omission notice (special message for browser_execute screenshots).
  • After gating, extract only model-supported media to separate messages when providers don’t support media in tool results.
  • Updated tests for vision vs. non-vision behavior; removed outdated screenshot media regression test.

^{Written for commit ee1f973. Summary will update on new commits.}

[cubic-dev-ai]

No issues found across 2 files

_{Re-trigger cubic}