deepseek-v4-flash is text-only in models.dev but DeepSeek's official API supports image input

Problem: When routing a request with an image to opencode/deepseek-v4-flash (via the opencode gateway, e.g. from Claude Code), it fails with "this model does not support image input".

Root cause: In anomalyco/models.dev (dev branch), models/deepseek/deepseek-v4-flash.toml declares text-only modalities:

[modalities]
input = ["text"]
output = ["text"]

DeepSeek's official API for deepseek-v4-flash supports image input (vision). Please update the modality declaration to:

[modalities]
input = ["text", "image"]
output = ["text"]

Also worth checking: deepseek-v4-pro.toml and deepseek-chat.toml have the same input = ["text"] declaration — they may also support vision upstream.

Related: #37879, #33199