lecture-python.myst: bake env + deploy CLI into the RunsOn GPU AMI

[mmcky]

#92 Phase 2, the lecture-python.myst (GPU/RunsOn) half of pre-baking: bake miniconda + env + pinned jax[cuda13]/numpyro + the preview-deploy CLI into the RunsOn AMI. The repo stays on RunsOn — CPU migration was evaluated and rejected in PROJECT-OPTIMIZE-PREVIEWS.md. Conda 102s + JAX 55s measured per PR are the target.

• Bake the AMI with an env_hash drift guard
• Per QuantEcon/infrastructure#4, the baked CLI is now wrangler, not netlify-cli
• Flip ci.yml to setup-environment's container-mode path — verify the AMI marker file (/etc/quantecon-container) exists first (Decision 3 of Optimize preview builds: phased plan for the fastest PR previews (tracking) #92)
• Staged rollout via workflow_dispatch