feat: Add deepseek-v4-flash-vision-exp to DeepSeek provider and organize README structure

[Haicaji]

What does this PR do?

1. organize README structure to avoid multiple README files in the project root

2. Closes [FEATURE]: Add deepseek-v4-flash-vision-exp to DeepSeek provider #80 — Add deepseek-v4-flash-vision-exp to the DeepSeek provider to support multimodal (image) capabilities.

Changes

1. packages/cyberstrike/src/provider/provider.ts

   • Added isVisionModel() helper that recognizes deepseek-v4-flash-vision-exp.
   • Forced capabilities.input.image = true for this model in both the models.dev database path (fromModelsDevModel) and the config-override path, so image uploads work even if models.dev / a stale local cache hasn't tagged the model with image modality yet.

2. packages/cyberstrike/src/provider/transform.ts

   • Clarified that the existing deepseek-v4 reasoning-effort branch (returns high/max) also covers the multimodal deepseek-v4-flash-vision-exp. No logic change needed — reasoning variants, thinking mode, and reasoning_content interleaving are inherited from the existing DeepSeek V4 handling.

3. Tests

   • Added deepseek-v4-flash-vision-exp to the models.dev fixture (test/tool/fixtures/models-api.json) with modalities.input: ["text", "image"].
   • test/provider/provider.test.ts: verifies the model loads from models.dev with image input / reasoning / interleaved reasoning_content capabilities and high/max variants, and that image capability is preserved even when config omits modalities.
   • test/provider/transform.test.ts: verifies variants() returns high/max reasoning_effort for the vision model.

Behavior

• Selecting deepseek-v4-flash-vision-exp in the DeepSeek provider now allows attaching images.
• Images are sent via the existing @ai-sdk/openai-compatible path as OpenAI-compatible image_url content blocks, which the DeepSeek API supports (base64 data URL / external URL).
• Thinking mode and reasoning_effort (high/max) variants work like the rest of the DeepSeek V4 family.

Type of change

• Bug fix
• New feature / agent
• Security tool / MCP server / Bolt plugin
• Agent skill / knowledge base
• UI / TUI improvement
• Documentation
• Refactor / performance
• CI / infrastructure

Security impact

• This PR adds or modifies tool execution (shell, file, network)
• This PR changes agent permissions or scope
• This PR modifies authentication / authorization logic
• This PR has no security impact

How did you verify it works?

To verify manually:

1. Add a DeepSeek API key (DEEPSEEK_API_KEY).
2. Select model deepseek-v4-flash-vision-exp from the DeepSeek provider.
3. Attach an image and confirm it is sent as an image_url content block and the model describes it.

Checklist

• bun turbo typecheck passes
• Tested locally with at least one LLM provider (unit tests)
• PR is focused on a single change
• No secrets, credentials, or API keys in the diff
• Breaking changes are documented (if any) — none

[github-actions]

Hey! Your PR title Add deepseek-v4-flash-vision-exp to DeepSeek provider and organize README structure doesn't follow conventional commit format.

Please update it to start with one of:

• feat: or feat(scope): new feature
• fix: or fix(scope): bug fix
• docs: or docs(scope): documentation changes
• chore: or chore(scope): maintenance tasks
• refactor: or refactor(scope): code refactoring
• test: or test(scope): adding or updating tests

Where scope is the package name (e.g., app, desktop, cyberstrike).

See CONTRIBUTING.md for details.

[github-actions]

Thanks for your contribution!

This PR doesn't have a linked issue. All PRs must reference an existing issue.

Please:

1. Open an issue describing the bug/feature (if one doesn't exist)
2. Add Fixes #<number> or Closes #<number> to this PR description

See CONTRIBUTING.md for details.