feat(e2e): make tests machine-selectable via REQ + feature markers

[nolte]

Summary

Make the E2E suite machine-selectable so a code/feature change can be mapped to the affected tests without hand-marking 700+ functions. Adds two auto-derived pytest marker axes (REQ + semantic feature), enforces marker registration with --strict-markers, and lifts each test's docstring TC-ID into the junit tc_id property. Also delivers a core-function coverage audit and removes a stale spec duplicate.

Changes

• Add auto-derived req<NNN> marker (from the test_req<NNN>_*.py file name) and a semantic feature axis (plant/watering/nutrient/harvest/calendar/journey) opted into via a module-level FEATURES tuple; a FEATURES value outside the registered set is a hard collection error (typo guard).
• Register all REQ + feature markers in conftest.py::pytest_configure and enable --strict-markers via a new tests/e2e/pytest.ini.
• Record each docstring TC-ID into the standard junit user_properties channel (tc_id) for downstream tooling.
• Annotate the 15 core-function test modules with their FEATURES.
• Add tests/e2e/README.md documenting the selection workflow.
• Add .audits/e2e-core-function-coverage/2026-07-14-core-function-coverage.md: spec-vs-implementation coverage map for the five core functions with named gaps.
• Remove the stale spec/test-cases/TC-REQ-001.md duplicate; canonical source is spec/e2e-testcases/.

Linked issues

Refs spec/e2e-testcases/ (coverage audit basis; removed duplicate spec/test-cases/TC-REQ-001.md)

Testing

• pytest --collect-only per axis: -m req004 -> 98, -m watering -> watering-log + journey, -m harvest -> REQ-007 + REQ-008, -m 'nutrient and smoke' -> 21; full suite 702 collected, --strict-markers breaks nothing.
• R4 verified end-to-end: junit <property name="tc_id" value="TC-REQ-004-W001"> emitted; no-op when no TC-ID.
• bash scripts/run-e2e.sh --smoke (full containerised stack): configfile: pytest.ini, 188 selected -> 141 passed, 42 skipped, 5 xpassed, 0 failed (exit 0).
• ruff: no new findings (working tree == HEAD baseline of 35 pre-existing).

Risk / rollout notes

Low. Test-tooling only — no backend/frontend/app source touched; the marker derivation is additive and the smoke run confirms no test regressed. --strict-markers is safe: the only decorator markers in the suite are the four already-registered ones. Coverage gaps are documented, not closed (separate follow-ups).