test: cover peak shaping strategies with data driven cases

[andig]

Fix #100, depends on #102

Follows up on #96 (comment): the peak shaping coverage lives in the existing data driven harness instead of a feature specific test module.

• three test cases, one per strategy: attenuate_demand_peaks levels grid import, attenuate_feedin_peaks moves charging into the solar hump, attenuate_grid_peaks does both
• a test case may now set "strict": true, which compares grid import, grid export, and the battery schedules with numpy.allclose on top of status and objective value
• strict is needed here because peak shaping only decides between cost neutral schedules. Without it the three cases pass unchanged when peak shaping is removed, the objective value differs only in the eighth digit
• existing cases stay non strict, several of them have equally optimal alternatives whose schedule comparison would be arbitrary
• testbench.py update keeps the strict flag when it refreshes a case

🤖 Generated with Claude Code

[andig]

@ekkea hope this is a better direction

[andig]

Added test_cases/026-attenuate-feedin-peaks-below-cap.json.

The three strategy cases pin the absolute peak; this one pins the shape of the profile below the peak. The surplus of the first step is far beyond c_max, so the export peak is fixed there and the remaining 170 Wh can be charged anywhere in the following steps without changing it — a schedule that only caps the peak has a free choice there, and the strict comparison pins the leveled one.

It fails on current main and needs #102:

• without the ramp penalty: grid_export [950, 50, 50, 50, 80]
• with it: grid_export [950, 57.5, 57.5, 57.5, 57.5]

🤖 Generated with Claude Code