feat: add DeepSeek V4 Pro and Flash models

[roomote-v0]

Related GitHub Issue

Closes: #12174

Description

This PR attempts to address Issue #12174 by adding the deepseek-v4-pro and deepseek-v4-flash model definitions as specified by @designgears. Feedback and guidance are welcome.

Key changes:

1. Added deepseek-v4-pro and deepseek-v4-flash to packages/types/src/providers/deepseek.ts with thinking mode (preserveReasoning), vision support, and prompt caching enabled.
2. Updated thinking mode detection in src/api/providers/deepseek.ts to use the preserveReasoning flag from ModelInfo instead of hardcoding the model name check (modelId.includes("deepseek-reasoner")). This makes it future-proof for any new thinking-capable models.
3. Added tests for the new v4 models covering model info, thinking mode enablement, and reasoning content streaming.

Test Procedure

• Ran cd src && npx vitest run api/providers/__tests__/deepseek.spec.ts -- all 28 tests pass (1 skipped is pre-existing).
• All lint and type checks pass via the pre-push hook (turbo lint and turbo check-types).

Pre-Submission Checklist

• Issue Linked: This PR is linked to an approved GitHub Issue.
• Scope: Changes are focused on the linked issue.
• Self-Review: I have performed a thorough self-review of my code.
• Testing: New tests have been added for the v4 models.
• Documentation Impact: No documentation updates are required.
• Contribution Guidelines: I have read and agree to the Contributor Guidelines.

Additional Notes

• Pricing values for v4 models are estimated. They should be verified against the official DeepSeek API pricing page once confirmed.
• The deepSeekModels object uses as const satisfies Record<string, ModelInfo>, so the handler code casts modelInfo to ModelInfo to access the optional preserveReasoning property safely.