Allow passing custom provider options to opencode run

[AurelienRibon]

The goal of this PR is to provide users of opencode run the flexibility to set and use every possible provider option proposed by the AI SDK.

This is a compromise with #3475, in which I proposed a dedicated argument for an openai-specific property.

This new approach is more technical (users need to check with Vercel docs for what is available), but has the large advantage to deport the maintenance from opencode devs to vercel devs:

• new models will be customizable without having to their properties to the code
• no headache to find how to generalize a property (like reasoning effort) to all models
• no arbitrary choice to make (mapping of reasoning effort from gemini's numbers to openai's enum)

Of course, dedicated arguments for important parameters will always be welcome, but at least now there will always be a workaround for missing arguments.

Does it work?

The best way to check that is to pass an invalid value and see if Vercel's AI SDK shows the correct error.
Here I tested it with reasoningEffort for openai models.

Note

Aiden, of course this is a proposal, feel free to discard it if that's not the way Opencode wants to go.

[rekram1-node]

I feel like this is a bit messy, you can also solve this currently using OPENCODE_CONFIG_CONTENT

OPENCODE_CONFIG_CONTENT="{"$schema":"https://opencode.ai/config.json","provider":{"openai":{"models":{"gpt-5":{"options":{"reasoningEffort":"high","textVerbosity":"low","reasoningSummary":"auto","include":["reasoning.encrypted_content"]}}}}}}" opencode run 'hello how are you'

Neither of these are great tbh but passing json through args feels messy and if peeople want to do that advanced behavior they could use the existing flag

[AurelienRibon]

Thanks, I was not aware of this flag. And I see it's doing a mergeDeep, so no need for this PR indeed, closing.

[rekram1-node]

I think the same thing applies for the timeout one too