[Feature Request] Support custom/static headers for LLM providers (needed for OpenCode Go)

Is your feature request related to a problem? Please describe. OpenCode Go (https://opencode.ai/zen/go/v1) requires an x-opencode-session: <stable id per conversation> header on all inference requests — requests without it error since 2026-09-06 (per vendor notice). {TOOL} does not send this header and offers no way to configure custom headers per provider, so any OpenCode-Go-backed provider configured in Vane will fail.

Describe the solution you'd like

• Provider-level headers support in the config/settings schema (static key-value pairs)
• Optionally: a substitution variable (e.g. {conversation_id}/{session_id}) that Vane fills in automatically per conversation/session, so the header stays stable across a session as required by OpenCode Go
• Alternatively (or additionally): add OpenCode as a first-class/built-in provider in Vane, so the session-header requirement is handled natively instead of relying on generic header support
• Either approach also future-proofs the provider integration against other vendors introducing similar correlation-header requirements

Describe alternatives you've considered No workaround — OpenCode Go cannot currently be used as a provider in Vane.

Additional context Same problem was reported for BrowserOS: https://github.com/browseros-ai/BrowserOS/issues/2543 — provider headers got merged there, but automatic session-id substitution is still open.