cli: fail fast when explicit model provider is not configured

[caozuoba]

Summary

This PR adds a fail-fast precheck in CLI startup to catch explicit model/provider mismatches early.

Before this change, configs like model=anthropic/... with only providers.zhipu.apiKey set could fail later at runtime with a less actionable provider error.
Now the CLI exits early with a clear configuration message.

What changed

• Added _ensure_explicit_model_provider_configured(config, model) in nanobot/cli/commands.py.
• Wired the precheck into _make_provider(...) before provider construction.
• Added tests in tests/test_commands.py for:
  • explicit provider mismatch -> exits early
  • explicit provider match -> passes
  • model without explicit provider prefix -> no strict block
  • _make_provider fail-fast behavior

Why this approach

• Keeps behavior unchanged for normal fallback flows.
• Adds guardrails only when users explicitly select a provider-prefixed model.
• Improves startup error clarity and reduces support/debug time.

Validation

Automated tests

python -m pytest -q tests/test_commands.py
python -m pytest -q

A/B manual verification

Use your local launcher (nanobot, nb, or a full executable path).
Below, <nanobot_cmd> means whichever command works in your environment.

Scenario A (guard case): explicit provider mismatch should fail fast

Config setup:

• agents.defaults.model = anthropic/claude-opus-4-5
• only providers.zhipu.apiKey is set

Command:

<nanobot_cmd> agent -m "hi"

Expected:

• immediate config error:
  • Model 'anthropic/...' requires provider 'anthropic'
  • prompt to set providers.anthropic.apiKey or switch model

Scenario B (happy path): matched provider should proceed normally

Config setup:

• agents.defaults.model = glm-4.7-flash
• providers.zhipu.apiKey is set

Command:

<nanobot_cmd> agent -m "Please reply with exactly: startup test passed" --logs

Expected:

• normal response from model (for example: startup test passed)

Notes

• No API/schema changes.
• No behavior change for users who do not use explicit provider-prefixed models.