[FEATURE]: Let agents propose executable action branches

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

For ambiguous tasks, an agent currently has to either choose an execution strategy silently or ask the user a generic question.

I propose an optional “action branches” interaction:

• The model decides whether branching is useful. Routine tasks continue normally.
• When needed, it proposes 2–5 alternative execution approaches with concrete commands and short tradeoff descriptions.
• The user can select one approach, or type feedback in an always-visible input.
• Selecting an approach executes it through OpenCode's existing tool and permission system.

Sequential execution should be the default and stop after the first failed command. Parallel execution should only be available for independent, read-only commands, and must never bypass normal permissions.

This is intentionally coarser than plan mode: it asks the user to choose between execution strategies rather than approve every implementation step.

If there is only one reasonable approach, the agent should continue normally without opening the branch selector.

I have a working V2 Core/TUI prototype:
https://github.com/dttutty/opencode/tree/action-branches

I would like feedback on whether this interaction fits OpenCode before opening a PR.