Restore in-progress books

[vitofico]

Summary

Phase 0–1 of the refactoring study (docs/refactoring-study.md) — the safe, no-behavioral-change slice. The goal of this PR is to fix the one concrete crash bug, lay a test safety net over the previously-untested data-access layer, and land low-risk hygiene/dedup, without changing any runtime/IO/exit/logging contract. The riskier behavioral items are deliberately deferred (see below) because this was executed unattended and they need human sign-off.

Scope was reviewed by an external model (plan review) which recommended exactly this conservative partition for an unattended run.

What's in this PR

#	Change	Risk
0.1	Add pytest-cov (dev dep + [tool.coverage] + CI --cov reporting). No --cov-fail-under gate.	none
0.2/0.3	tests/conftest.py + characterization tests pinning current APIClient behavior (URL handling, success path, and the existing error-swallowing/dummy-token behavior)	none (tests only)
1.1	Bug fix: add params= kwarg to APIClient.get. API.get_device_data/get_weather_data already call client.get(path, params=...), which raised TypeError on every device-data/weather fetch. Covered by new regression tests.	low
1.4	Replace 3 production asserts (measured.py, pipeline.py) with explicit raise guards — assert is stripped under python -O and violates the repo's own ast-grep rule. Behavior-equivalent under normal execution.	low
1.5	Bump requests 2.31.0 → >=2.32.4 (CVE-2024-35195, CVE-2024-47081); drop redundant matplotlib/weasyprint from the dev group.	low
1.7	Extract the byte-identical _quarter_label (duplicated in pipeline.py and bundle.py) into a single config.quarter_label.	low

Verification

• uv run pytest → 161 passed, 2 deselected (was 150; +11 new tests, no regressions).
• uv run pre-commit run --all-files → ruff check, ruff format, and ast-grep all pass.
• Coverage now reported in CI (was 0% on src/api; src/utils/data_joiner.py and timeseries_generator.py remain ~0–2% — see deferred work).

Deliberately deferred — needs human sign-off

These were identified and intentionally NOT done because they change runtime behavior in zero-coverage areas where production env/consumer assumptions can't be verified unattended. Each is documented in docs/refactoring-study.md:

• 2.2 Honest IO — make authenticate()/get() raise instead of returning a dummy token / {"data": []}. ~20 consumers assume a dict and never catch; no mid-run degrade path exists today. The characterization tests in this PR pin the current behavior precisely so this change can be made safely as a reviewed follow-up.
• 2.1 Config consolidation — collapsing config.py/dps_config.py switches DPS credentials from '' → admin/password; committed Argo manifests don't inject API_USERNAME/API_PASSWORD, so DPS may currently run silently offline. Needs prod-env confirmation.
• 2.5 Exit/error contract & 1.2 false-success exits — some empty-DataFrame returns are legitimately "no data", others mask failures; misclassifying breaks pvbq's relaxed-completeness handling and changes exit codes CI/Argo depend on.
• 1.3 DPS type=bool flag — fix requires editing live Argo CronWorkflow manifests in lockstep.
• 1.6 RUF013 + --unsafe-fixes — repo-wide implicit-Optional autofix is too much unreviewed annotation churn for an unattended run.
• 1.8 / 1.9 / 2.3 / 2.4 — config/data-key dedup, logging rewiring, dead-code deletion (CatalogService, SiteReport, …), logging-bootstrap unification: each touches import-time timing, operational log/stdout contracts, or needs a repo-wide "is this really dead / imported-by-name?" audit.
• Phase 3 (decomposing the 1700/1015/748-line functions): separate PRs once the safety net and seams are in place.

Reference

Full analysis, target architecture, and prioritized roadmap: docs/refactoring-study.md.