Refresh default model catalog for Claude Opus 5 and GPT-6

[edwin-zvs]

Summary

• Adds Claude Opus 5 (claude-opus-5) to the Anthropic completions block in the curated model registry (crates/protocol/src/slash.rs), and advances the Claude-subscription sonnet/opus short-alias expansion (resolve_claude_oauth_model) to point at claude-sonnet-5 / claude-opus-5 instead of claude-sonnet-4-6 / claude-opus-4-8.
• Adds the new GPT-6 generation (gpt-6-astra flagship, gpt-6-sol, gpt-6-luna, plus the gpt-6.1-sol refresh) to both the codex-oauth and openai completions blocks.
• Advances the default model in all five independent per-provider fallback ladders so the new flagships are picked when no model is pinned:
  • crates/adapter-smith/src/agent.rs::default_auto_detect_spec() — ANTHROPIC_API_KEY → claude-opus-5, OPENAI_API_KEY → gpt-6-astra.
  • crates/adapter-smith/src/interactive.rs::default_monitor_spec() — anthropic monitor tier → claude-sonnet-5.
  • crates/daemon/src/config.rs::BUILTIN_TARGETS — anthropic/openai default_model updated; doc-comment table updated.
  • crates/daemon/src/availability.rs::smith_auth_methods() — anthropic/openai env-key defaults and the Claude-subscription default_model updated.
  • crates/adapter-smith/src/title_mode.rs::pick_default_spec_str() (cheap/mini tier) is intentionally untouched — no confirmed cheap GPT-6 tier exists yet.
• Updates docs/model-routing.md and docs/smith.md to match the new defaults and example model ids.
• Fixes three pre-existing test fixtures that hardcoded the old defaults and would otherwise drift: claude_oauth.rs's own resolve_model alias test, the daemon router's claude_oauth_published_aliases_expand_to_concrete_models, and slash.rs's models_for_provider("anthropic") ordering assertion.

All additions are additive — existing concrete model ids (claude-opus-4-8, claude-sonnet-4-6, gpt-5.6-sol/terra/luna, gpt-5, etc.) remain in the registry, so any session that already pinned one of those keeps working. Only the default/fallback values move forward.

Google/Gemini: no code change included. Gemini 4 ("Argon") has no confirmed, publicly callable API model id yet (still limited-tester) — the existing gemini-3.8-flash / gemini-3.1-pro entries remain the newest actually-callable Gemini models. Flagging this explicitly so the omission reads as a deliberate finding, not a miss.

Test plan

• cargo build (debug profile) succeeds in the worktree — only pre-existing unrelated dead-code warnings, no new warnings/errors from this change.
• cargo test --workspace (full, unfiltered suite) passes clean on a committed (non-dirty) tree: 0 failures.
• Binary produced: target/debug/construct — this is the only workspace binary, and it contains every file this PR touches (crates/protocol, crates/adapter-smith, crates/daemon all link into it).

🤖 Generated with Claude Code