Opencode should fall back to cached well-known config when endpoint is temporarily unavailable #10930

Description

If a user has a wellknown credential configured, Config.get() fetches <url>/.well-known/opencode during startup. If that endpoint is temporarily unavailable (e.g. 500, network hiccup, timeout), Opencode fails to start because config loading hard-throws.

This makes org/enterprise setups brittle: a transient config outage bricks the CLI/agent even if local/project config could still be used.

Suggested fix

• Cache the last successful well-known config payload locally (e.g. XDG cache).
• On fetch failure, fall back to the cached config (with a warning) instead of hard-failing.
• Add a configurable fetch timeout (e.g. OPENCODE_WELLKNOWN_TIMEOUT_MS).

Steps to reproduce

1. Configure a well-known credential (e.g. via opencode auth login <url>).
2. Ensure <url>/.well-known/opencode becomes unavailable (500 / offline) after a successful run.
3. Start Opencode again.

Expected

Opencode starts and uses the cached well-known config (and logs a warning).

Actual

Opencode throws failed to fetch remote config from <url>: <status> during config load and exits.

Environment

• OS: Windows 11 (also affects any OS on transient network failures)
• Terminal: Windows Terminal