[ArtyMend07]

Summary

• Resolves every neural population by FlyWire cell type instead of a fixed count per eye. The hallway eye is now LPLC2 plus LC4, every cell of each type.
• Lets the fly watch Foxy through the tablet. The camera feed is captured and the camera map is clicked from the screen; figure channels (LC9, LC31a) drive DNp09, which raises the tablet and follows him, and loom channels (LPLC2, LC4) drive DNp04, which drops the tablet and closes the left door. The GABAergic camera inhibitors are driven while the tablet is seen. The tablet drive is recorded in the trace and replayed, and the panel shows the camera feed, DNp09 and DNp04 on the timeline.
• Adds scripts that measure the tablet pathways and search between cell types.
• Moves the source into the flynaf package under src/, declares it in pyproject.toml and runs the tests under pytest, with a connectome marker for the tests that need the FlyWire data (ADR 0027).
• Starts the night only after the network has settled, reads every timestamp from one high-resolution clock (ADR 0028), measures the screen in physical pixels and writes each run to its own log file.
• Fixes the fly reopening the left door with Bonnie behind it. Behind a closed door the eye now reads the window rectangle instead of the door box, the closed-door bank keeps only the brightest lit frame, the threshold went from 40 to 30, and a look only clears a door when the light was confirmed on screen (ADR 0011 v1.4, experiment 14).
• Cuts the browser title bar off the panel, rounds its corners, and moves a narrow game window to the left edge so the full panel fits beside it (ADR 0021 v1.7).
• Updates the README and ADRs, removes the roadmap and bumps the version to 0.3.0.

Test plan

• ruff check .
• Fast suite (pytest -m "not connectome"): 251 passed
• Live night: Bonnie behind the closed left door read 95 against 30 and the hold was renewed, the door was released only when the window read 0; Chica read 482 behind the right door
• Live docking: the game moved to the left edge and the panel opened beside it at full height
• test_inhibition.py and test_pathways.py (connectome marker) have not been rerun after the package move
• The right window rectangle has not been measured on Chica yet (this chicken didn't show up on my runs LOL), it still uses the old box