feat(t2-3): replace global config singleton with local TycoonConfig construction

[JesuFemi-O]

Summary

• Removes the from tycoon.config import config global singleton import from four ingestion command modules: sources.py, run_all.py, sync_cmd.py, and explore.py
• Each command now constructs TycoonConfig(project_root=_find_project_root()) at invocation time, so every CLI call gets a fresh, correctly-rooted config without shared mutable state
• sources._require_project() now returns the TycoonConfig it builds so all helpers receive it as an explicit argument; _maybe_auto_scaffold gains a cfg keyword parameter
• Tests drop monkeypatch.setattr(mod, "config", ...) in favour of monkeypatch.chdir(tmp_path) — simpler and more faithful to real execution

Files to focus on

Production changes (2 patterns, 4 files):

• src/tycoon/commands/sources.py — largest change; _require_project() return type + _maybe_auto_scaffold signature
• src/tycoon/commands/run_all.py, src/tycoon/commands/sync_cmd.py, src/tycoon/commands/explore.py — mechanical: swap import, add cfg = TycoonConfig(...) at top of command function, replace config.X → cfg.X

Test changes (drop singleton patching):

• tests/test_sources.py, tests/test_explore.py, tests/test_sync.py, tests/test_notify.py, tests/test_templates_e2e.py

Test plan

• uv run ruff check src tests — clean
• uv run pytest -q — 690 passed, 3 skipped

Part of milestone M2 (#83). Stacked on feat/m2-t2-2.

---

^{Need help on this PR? Tag @codesmith-bot with what you need. Autofix is disabled.}

Closes #94

[gemini-code-assist]

Caution

The consumer version of Gemini Code Assist on GitHub has been sunset. All code review activity has officially ceased.

[db-tycoon-stephen]

Cleanest PR of the stack. The conversion is complete and consistent across all four command modules, and the test churn is genuinely just cleanup — I checked the three test_explore.py tests that lost their monkeypatch.setattr(..., "config", ...) binding, and all three already had monkeypatch.chdir(tmp_path), so they still assert what they claim rather than passing for the wrong reason. No assertions were dropped anywhere in the -72 lines.

One dead call: cfg.reload() at sources.py:494 is now a no-op. Under the singleton it was load-bearing — it refreshed shared state for anything running later in the same process. Now cfg is function-local and never read again after that line (add_source only calls success(), _maybe_install_*(), and next_steps() before returning), so it just re-reads a file and throws the result away. Safe to drop.

Two structural notes, neither blocking:

TycoonConfig(project_root=_find_project_root()) is now duplicated at four call sites, importing a private helper across module boundaries. sources.py already has the better shape — _require_project() returning the cfg. Promoting that into config.py as something like load_config() or TycoonConfig.from_cwd() would give one public entry point and stop _find_project_root leaking out of the module that owns it.

The singleton is still live in transform.py, db.py, and status.py — test_templates_e2e.py::_rebind_config still has to patch transform_mod.config and cfg_mod.config, which is the tell. Perfectly reasonable to land the migration in slices, but worth an issue tracking the remaining modules so the codebase doesn't settle permanently into two competing patterns.

[db-tycoon-stephen]

Requesting changes on the items in the comment above: drop the now-dead cfg.reload() at sources.py:494, and promote the duplicated TycoonConfig(project_root=_find_project_root()) into a public factory rather than importing a private helper across four modules.

Also worth opening an issue for the singleton still live in transform.py/db.py/status.py so the two patterns don't become permanent. Strongest PR of the stack otherwise — the test cleanup checks out.

[db-tycoon-stephen]

All three addressed. load_config() is the right shape and the docstring pointing people away from _find_project_root is a nice touch — no imports of it left in commands/. You also caught the second dead reload() in remove_source that I missed. Thanks for filing #175 for the remaining modules.