fix(photonics2d): honor lambda1/lambda2 passed per call

[mkeeler43]

Fixes #274.

simulate(design, config={"lambda1": ..., "lambda2": ...}) ignored the wavelengths. The solver reads self.omega1 / self.omega2, which v0 only computes in __init__; all other conditions are read from the dict _setup_simulation returns, which is why config= works for them.

v1 now overrides _setup_simulation to recompute both frequencies from the merged conditions. simulate and optimize both call it.

v0 is unchanged so that published v0 results stay reproducible.

Verified

Simulating the first five test-set designs under their own conditions, with one problem instance and the conditions passed per call:

row	stored objective	before	after
0	4.7460	0.2199	4.7460
1	5.7242	1.0781	5.7242
2	6.0318	0.3333	6.0318
3	5.7660	1.9619	5.7660
4	5.9565	1.3911	5.9565

The same design at the default wavelengths still gives 0.2199, so the fix changes the result only when the wavelengths differ.

Tests

• test_simulate_honors_wavelengths_passed_per_call: wavelengths passed per call give the same result as wavelengths passed to the constructor. Fails on main.
• test_dataset_row_reproduces_its_own_objective: simulating a dataset design under its own conditions returns the stored objective. This would be worth adding for the other problems too; beams2d passes it to within 1e-5.

[SoheylM]

The wavelength fix looks correct. Recomputing omega1 and omega2 from the merged per-call configuration covers simulate, optimize, and render. The new regression tests also pass locally: 15 passed.

The remaining actionable blocker is formatting: GitHub Actions’ Ruff 0.16.3 reports that tests/test_photonics2d.py would be reformatted. Please run ruff format tests/test_photonics2d.py and push the result.

The Linux 3.10 failure appears unrelated to this change: two Podman container tests fail because newuidmap is unavailable. A new push will rerun CI against the current base. Once formatting passes and the required checks are green, I’m happy to approve.