Add out-of-the-box demos with quadruped locomotion RL demo

[myurasov-nv]

Summary

Adds an out-of-the-box demos mechanism to Isaac Automator, plus the first demo: Quadruped Locomotion RL.

• New --demos deploy option (all clouds): comma-separated demo list, or no (default).
• Demo registry + dependency auto-enable: selecting a demo automatically enables the apps it depends on (using their default git refs) if they were left off, so a demo deploys with a single flag. Unknown demo names fail fast.
• New demos Ansible role: installs, per selected demo, an executable launcher under ~/.local/share/isaac-automator-demos/ and a double-click desktop shortcut.
• quadruped-locomotion demo: launches RSL-RL training of Isaac-Velocity-Flat-Anymal-D-v0 in Isaac Lab, rendered live in the Isaac Sim viewport. Tunable via HEADLESS / NUM_ENVS / MAX_ITERATIONS env vars. Depends on Isaac Sim + Isaac Lab (auto-enabled).

Implementation

• src/python/config.py - --demos default + demo registry (requires per demo).
• src/python/deploy_command.py - --demos CLI option.
• src/python/deployer.py - resolve_demos(): validate names, auto-enable required apps.
• src/ansible/inventory.template - pass demos to Ansible.
• src/ansible/roles/demos/ - new role (launcher + desktop shortcut), wired into isaac-workstation.
• README.md - documents --demos and the Demos section.

Tests

• src/tests/deployer.test.py - 4 unit tests for the resolver (no-op, unknown-demo exit, auto-enable, explicit-ref preserved).
• src/tests/e2e/verify.py - new check_demos verifier check.
• src/tests/e2e/lifecycle.py - --demos passthrough.

Verification

Full source deploy on an AWS A10G workstation (Isaac Sim v6.0.0 + Isaac Lab release/3.0.0-beta2) with --demos quadruped-locomotion:

• PLAY RECAP: failed=0; demos role installed launcher + shortcut.
• e2e verifier: 7 passed / 0 failed (incl. Demos install: quadruped-locomotion).
• Launched the demo end to end: a full 300-iteration RSL-RL training run on the ANYmal-D flat-locomotion task (64 envs), with live reward/metric logging; viewport renders the quadruped grid.

Note: on the pinned release/3.0.0-beta2, ./isaaclab.sh has no train subcommand, so the launcher uses the -p scripts/reinforcement_learning/rsl_rl/train.py form (the deprecation notice it prints refers to a newer Isaac Lab).