Capture an LED from several cameras at once (--device 1,2)

[gunicsba]

--device now takes a comma separated list of cameras, and every LED is lit once and looked for in all of them at the same time:

marimapper wled --server 192.168.1.50 --device 1,2
marimapper_check_camera --device 1,192.168.1.20:4747     # webcams and network cameras can be mixed

Each camera is its own view of the scan, numbered in the order they are listed. The reconstruction already treats every view as a separate camera position sharing one camera model, so with identical cameras a scan gives you one view per camera for the price of one.

This builds on #90 (it uses its address parsing), so the first commit here is that pull request's. The change for this PR is the second commit.

What changes

• find_leds and enable_and_find_leds look for one LED in every camera. The single camera functions are now thin wrappers around them, so there is one code path.
  • A camera that always sees some light is skipped for that LED (after the usual 3 seconds), the other cameras carry on.
  • The response time that feeds the timeout counts the slowest camera that saw the LED. A camera that never sees an LED would otherwise make the timeout grow by 50% every time.
• Every camera finishes its own view, as done, or deleted if it moved or saw nothing. The scanner waits for all of them and moves on by one view per camera. A scan counts as successful if at least one view is. A camera that detected nothing now ends its view too (deleted); before, nothing was sent in that case, so the scanner would have kept waiting.
• open_cameras opens every webcam with the same capture method and checks that each delivers a picture. Methods number the cameras differently (on Windows, DirectShow and Media Foundation list them in a different order), and the old per camera fallback could mix the two, so a list like 2,3 opened the wrong camera for the second entry. If no method works the error says why.
• The 2D files were named after the second, so two cameras finishing together overwrote each other's file. A suffix is added when the name is taken.
• marimapper_check_camera opens a window per camera, and the README has a short section on it.

Testing

• Unit tests use a fake scene (cameras, an LED backend) to check which camera sees which LED, a camera that is blind or always sees light, the timeout not inflating, one result per camera, and the file names. open_cameras is tested with a fake OpenCV, including the case where one method can open only some of the cameras. The scanner side is tested with a scripted queue of events. All pass, flake8 is clean, and the existing tests pass.
• On real hardware: three webcams at once (two identical USB webcams on separate ports and a laptop webcam), four scans with the WLED backend. That gave 10 views, and the live reconstruction placed 191 of 200 LEDs. One of the scans produced two files in the same second, which is the naming case above.

Notes

• Two identical webcams on the same USB port or hub may not stream together (the second delivers no frames). Separate ports fixed it here. The README says so.
• Every camera gets the same --exposure, there is no per camera setting.
• This touches the same part of scanner.py and the README as two other pull requests, Show per-camera progress with detection rates and gaps while scanning #92 (progress display) and Add --capture_only and a marimapper_reconstruct command #93 (--capture_only). They're independent, but whichever lands second needs a small rebase. I tried all three merged together: they combine and all 74 tests pass.

🤖 Generated with Claude Code