feat: tutorial dataset system with xarray.tutorial-like API

[pgierz]

Summary

• Add tutorial loader with entry-point model discovery (pycmor.tutorial.open_dataset())
• Add stub data generator from YAML manifests
• Add data fetcher with pooch-based caching
• Add AWICM-RECOM as built-in tutorial dataset with registry and stub manifest
• Add tutorial unit tests
• Add documentation for model fixture development and plugin architecture

Context

Part of the future/workbench disentanglement. Provides an xarray.tutorial-like interface for loading example climate model datasets. Users and tests can import pycmor.tutorial and load registered model data without manual path wrangling.

Test plan

• CI passes on this branch
• tests/unit/test_tutorial.py passes
• Tutorial loader discovers registered datasets
• Stub data generation from YAML manifests works

🤖 Generated with Claude Code

[pgierz]

Merged into #253 -- these two features have circular dependencies (test-infrastructure needs tutorial datasets, tutorial tests need entry-point model discovery) and must ship together.