Add Molecular Devices SpectraMax Gemini EM backend

[brad-usredoxlabs]

Summary

Adds a Molecular Devices SpectraMax Gemini EM plate reader backend, including exports, backend tests, and user-facing documentation.

Validation

• python3 -m unittest pylabrobot.plate_reading.molecular_devices.backend_tests
• Direct Linux hardware smoke test over /dev/ttyUSB0: setup, status, temperature, open, close
• Direct Linux hardware validation over /dev/ttyUSB0: fluorescence, luminescence, TRF endpoint, TRF kinetic, emission spectrum, excitation spectrum, fill wellscan, and partial-region fluorescence

Hardware Results

• Fluorescence: 1 read, 8 x 12
• Luminescence: 1 read, 8 x 12
• TRF endpoint: 1 read, 8 x 12
• TRF kinetic: 2 reads, each 8 x 12
• Emission spectrum: 2 reads, each 8 x 12
• Excitation spectrum: 2 reads, each 8 x 12
• Fill wellscan: 9 reads, each 6 x 6, with scan point/x/y metadata
• Partial fluorescence B2:G7: 1 read, 6 x 6

nan values appeared as expected where the reader returned blanks.

Notes

Hardware validation was run on a Linux laptop directly connected to the SpectraMax Gemini EM through a USB-to-RS-232 adapter. ruff was not run locally because python3 -m ruff is not installed in this environment.

[rickwierenga]

Pushed 0d80c88 addressing four review findings:

1. State leak between reads — read_fluorescence now calls _set_wellscan_mode(False). Previously it was the only read method that didn't, so a prior read_fluorescence_wellscan would leave WELLSCANMODE=ON and a subsequent plain fluorescence read would inherit that state.

2. Silent focal_height ignore — read_fluorescence now raises NotImplementedError when focal_height is not None, matching the existing behavior of read_luminescence and read_fluorescence_wellscan. Tests that passed focal_height=0 to read_fluorescence have been updated to drop the argument, with a new test_fluorescence_rejects_focal_height covering the strict path.

3. Plate-specific wellscan defaults — read_fluorescence_wellscan now derives center_x/center_y from plate.get_item(0).location (same logic as _set_plate_region) when not supplied. The previous defaults (14.380, 20.235) were captured from one dev plate and silently wrong for any other geometry. The OEM-validated sequence test now passes those values explicitly to preserve the captured command stream.

4. Order-sensitive well comparison — _assert_full_plate and _get_well_region now compare by {id(well) for ...} set semantics. Previously wells != plate.get_all_items() used Python list equality, so a reordered full-plate list would fail the fast-path and either raise NotImplementedError (wellscan) or fall through to the rectangular check unnecessarily.

All 69 tests in backend_tests.py pass locally.

[rickwierenga]

@brad-usredoxlabs really good PR!

pushed some changes and converted the hello world from markdown to notebook so it's easier to run. could you please verify this works?

also renamed the non-standard-PLR methods to experimental_ because we want to figure out a good call+return type system for them. In this way, it is clear to users this function might change in the future while still being useful today.

[rickwierenga]

thanks and congrats on first PR!