[FEATURE]:Add an opt-in "data-sharing free-tier fallback" for subscribed users during usage-limit windows

Feature hasn't been suggested before.

• I have verified this feature I'm about to request hasn't been suggested before.

Describe the enhancement you want to request

Problem

Paid users (OpenCode Go subscription and/or Zen credit balance) who exhaust their 5-hour / weekly / monthly usage window currently have no $0 continuation path. The only ways to keep working are:

• Wait for the usage window to reset (minutes to days of downtime), or
• Use the paid Zen balance spilling (costs money), or
• Cancel the subscription entirely to return to the free tier — a destructive, all-or-nothing action.

There is no lightweight, temporary, opt-in way for a paid user to continue on free models without permanently changing their account status.

Suggestion

For subscribed accounts that reach their paid usage limit, offer an explicit, opt-in toggle such as:

"Temporarily route me to free models. In exchange, I agree that my request data may be used to train models. I understand this applies until my paid usage window resets."

Requirements it would need to be fair and safe:

1. Explicit consent screen stating exactly what data is shared and for how long.
2. Visible indicator in the app (e.g. a banner per request) when falling back to a free model, so the user always knows requests are now training data.
3. Automatic return to paid models the moment the paid usage window resets.
4. No effect on subscription status — staying subscribed, just temporarily routed to free models with consent.

Why it is fair

Free models exist for users who "pay" with their training data. Subscribed users currently "pay" only with money and are locked out of the data-for-tokens trade entirely. Giving paid users a time-boxed, informed-consent opt-in would let them choose temporary data-sharing instead of forced downtime or forced cancellation — with no change to the underlying economics, just more user agency and control.

Notes

• Precedent already exists: opencode ships free models explicitly labelled as "data used for training" (e.g. MiMo-V2.5 Free, Big Pickle, Hy3 Free, Nemotron Free). This extends that same transparent, consent-based model to a temporary fallback path for subscribed users.
• Alternative (even simpler): even a "manual consent before each free-model request during a window" flow would satisfy most users.