refactor(cli): extract interactive menu into cli/interactive.py with CliContext

[ArtVsMark]

Summary

Issue #121 Phase 2, part of the cli.py decomposition epic (#117). Phase 1 (rendering extraction) already merged. Branched from fresh main (#118-#120/#121-Phase-1 all merged).

Extracts the interactive-menu/prompt subsystem — _interactive_menu, _ask_bench_profile/_ask_micro_profile/_ask_number, _print_menu, _pick_path_via_dialog/_prompt_path/_resolve_cli_path_or_error, _BENCH_PROFILES/_MICRO_PROFILES — out of cli/__init__.py into a new leaf module cli/interactive.py, following the CliContext-based DI pattern #120 already proved (842/842 tests passed there with zero test changes).

What I verified before designing

Grepped every test file to find exactly which of these names are directly monkeypatched via cli.<name> (expecting effect to propagate through the call chain) vs. never patched: _ask_bench_profile, _ask_micro_profile, _pick_path_via_dialog, and _interactive_menu are; _print_menu, _ask_number, _prompt_path, _resolve_cli_path_or_error, and the two profile dicts are not. CliContext gains 7 new fields for exactly the directly-patched names; _print_menu/_prompt_path stay as ordinary same-module bare calls inside interactive.py since nothing independently patches them — confirmed, not assumed — and they still correctly propagate patches on their own dependencies (e.g. _prompt_path → ctx.pick_path_via_dialog).

The i18n-state scope decision

The issue's Phase 2 text groups these functions with _LANG/_MESSAGES/_LOCALE_MESSAGES/_t. I deliberately did not move the i18n state — main() does global _LANG; _LANG = args.lang, a real runtime mutation, not just monkeypatch plumbing. If _LANG physically moved to a new module, cli/__init__.py's re-export would become a one-time snapshot instead of a live alias, silently breaking monkeypatch.setattr(cli, "_LANG", "en") (an autouse fixture covering ~68 tests in test_cli.py) — the exact class of hazard #120 was designed to avoid, but worse here because it's real state, not just test plumbing. A Plan agent independently re-verified this before implementation and confirmed it as a hard blocker, not a cautious guess. _LANG/_MESSAGES/_LOCALE_MESSAGES/_t stay in cli/__init__.py, unmoved — treated as a settled architectural boundary, not a "later" TODO.

Result

Zero test files changed. The full suite (842 tests) passes unmodified.

This design was validated by a Plan agent before implementation (independently re-verified the monkeypatch inventory, the context-field-vs-bare-call rule, and the _LANG mutation hazard) and approved via plan mode before any code was written.

Test plan

• pytest tests/ -q — 842 passed, 3 skipped, 0 test files modified (deselected the 2 pre-existing unrelated failures already failing on main)
• ruff check src tests docs / ruff format --check src tests — clean
• mypy src/stepik_grader --ignore-missing-imports — clean
• Manual smoke: interactive menu (mode 0 exit via piped stdin), --mode 1 --file ... --output json, interactive-menu-driven mode 1 path prompt — all exercised the new wrapper → interactive.py → CliContext chain end-to-end
• from stepik_grader.cli import main, import stepik_grader.grader — facade re-exports intact
• Updated docs/architecture.md (module table + DAG) and docs/project-structure.md for the new cli/interactive.py leaf module and CliContext's extended field list

🤖 Generated with Claude Code