test(annual): land the heat pump tests stranded in PR #33

[dandwhelan]

Closes the CI coverage gap on the heat pump comparison. Supersedes #33.

Why not just merge #33

The heat pump feature shipped its production code — annual_heat.py (881 lines), annual_costs.py, and the annual.py / web_annual.py additions are all on main and running live on v712.23. Its tests never landed, and have sat in #33 since 2 August.

#33 can't be merged now to fix that. main's copies have moved a long way past it:

file	main vs #33
web_annual.py	+336 / −101
annual.py	+104 / −31
annual_heat.py	+1 / −4

That drift includes the upstream v8.49.0 merge (#46) and the battery-cycles fix (#47). A trial merge conflicts on four files, and resolving it would risk dragging production code back to 2 August to fix a test gap. So this takes the tests only, rebased on current main.

What's here

• test_annual_heat.py — 752 lines, a file main does not have at all
• make_heat_model, test_annual_heat_integration → test_annual_integration.py
• heat_results, test_web_annual_heat, test_web_annual_predheat_prefill → test_web_annual.py
• 4 suites registered in unit_test.py; subtractable added to the workspace dictionary

One deliberate change from #33

test_annual_heat_integration is registered through a run_annual_heat_integration_isolated wrapper rather than against the shared instance. #33 predates run_annual_integration_isolated, which exists because that test plans a year of sampled days without setting up the state it depends on — inheriting ambient state made it 13x slower (34s → 463s) without ever failing. test_annual_heat_integration is the same shape of test and wants the same treatment for the same reason.

The guards worth having

• test_annual_heat_integration pins the heat leg's extra no_pvbat import to exactly one day of heat and explicitly not the two-day figure — this is what catches the unbounded end_record regression, where run_prediction() returns running totals over the whole 48-hour plan rather than the end_record window.
• The model tests hold the identity that the modelled gas bill must not move when the heat pump is added, and that space heating and hot water divide by their own separate efficiencies.

Verification

Full quick suite 242 passed, 0 failed (3 new non-slow suites; annual_heat_integration is slow-marked). All four heat suites pass individually against current main. black and ruff --select=F401 clean.

🤖 Generated with Claude Code