Document and verify tutorial commands across backends and shells

[marpaia]

Tutorial readers can select CPU, Metal, or CUDA using equivalent commands and a shared shell guide. The guide covers device discovery, POSIX and PowerShell JSON quoting, paths with spaces, saved-parameter checkpoint resume, and Stop/restart. Tutorial, environment and viewer pages link directly to it. Resume instructions now match automatic restoration of the saved seed and parameters.

The regression harness extracts literal Markdown commands and runs real shells and CLI processes in a checkout path containing spaces. It verifies authenticated checkpoint provenance, resumed state equality, rejected parameter/source changes, unavailable-backend errors, and port reuse. A descendant-process regression keeps failed command tests from leaving child servers alive.

Validation: 10 tests passed on macOS across sh, Bash, Zsh and PowerShell, including native CPU and Apple M4 Max Metal trap execution, CPU checkpoint/resume equality, 12 live Stop/restart launches, link/flag/path checks and bounded child-process cleanup. Ruff, formatting and strict Pyright passed. Actual Windows CPU/PowerShell validation passed 4 tutorial/cleanup tests and 28 shutdown/console-interruption tests at implementation commit 215e87d; the final commit only records these results. No NVIDIA runtime execution is claimed. PowerShell JSON guidance requires 7.3+ with Standard argument passing; Windows PowerShell 5.1 and cmd are outside that guidance.

Originally reviewed against frozen marpaia/base-24, combining #17 and #22. Prerequisites were integrated before retargeting to master. Exact platform and command evidence is documented in docs/development/tutorial-command-validation.md.

Closes #24.

Independent review is complete against the original acceptance criteria and intended review base. The review report records per-criterion evidence, both corrected campaign findings, exact tested heads, landing order and platform limits.

Merged into master in the approved campaign order: #26, #27, #28, #31, #34, #36, #29, #30, #33, #35, #37, #32. The final tree preserves the reviewed implementation and integration regressions. Prerequisites were integrated with merge commits, and the frozen review bases were retained.

Landing validation: 534 Python tests passed (57 environment-dependent skips), 115 viewer tests passed, Ruff lint, strict type checks, viewer build/format checks, and actual Chromium feature/integration checks passed. Pyright retains the existing native-extension source warning. A Windows-only documentation link-test encoding failure was reproduced and fixed with an explicit UTF-8 read; the corrected final PR heads passed Windows CLI/shutdown checks and CUDA compilation. NVIDIA runtime execution remains unverified.