Support DeepSeek V4.1 Flash in the native DeepSeek provider

[iluv7]

I would like to contribute DeepSeek V4.1 Flash support to the native DeepSeekChatModel. It is currently listed under DashScope, but not in the native DeepSeek catalog.

My plan is to:

• Confirm the official API model ID, availability, and supported capabilities.
• Add the model card and adapt request parameters or message formatting where needed, reusing the existing DeepSeek implementation.
• Add tests for model discovery and any new behavior, including thinking and tool calling where supported.

Would this contribution be welcome? If the approach looks good, I will open a PR.

[qbc2016]

@iluv7 Thanks for looking into this! A PR would be welcome.

[iluv7]

After checking the official API docs and the existing implementation, I found that deepseek-flash already serves DeepSeek V4.1 Flash, so my initial description of missing native support was inaccurate. Sorry for the confusion.

The remaining gaps appear to be the model card's output limit (384,000 → 393,216), support for the none and low reasoning-effort values, and image capabilities on the legacy deepseek-v4-flash alias. I plan to narrow the PR to these corrections and focused tests, rather than add a duplicate model entry.

References: model names and aliases, API parameters and exact output limit.

Would this narrower scope be appropriate for a PR?