[codex] default to glm-5.1

[ChunkyMonkey11]

What changed

• Switched the NVIDIA default routing path to z-ai/glm-5.1.
• Kept glm, qwen, and smoke as explicit alternatives, with clearer tradeoffs in docs and CLI help.
• Bumped the package version to 0.1.2 and tagged the release as v0.1.2.

Why

• The previous default model had a smaller context window and could fail on long Claude Code prompts.
• GLM-5.1 gives a much larger context window, which makes the default onboarding and demo path more reliable.

Impact

• Fresh installs now default to the long-context NVIDIA model.
• Users still have access to slower/higher-reasoning and fallback profiles when they want them.
• The router now handles context overflow cases more gracefully and surfaces a clearer error when the prompt itself is too large.

Validation

• npm test -- tests/status.test.ts tests/profile.test.ts tests/claudia-config.test.ts tests/openai.test.ts
• npm run build