docs: examples, copy-and-fill templates, notebook tests; remove experiments/

[hinderling]

What

Replaces the experiments/ folder (migrated with full history to pertzlab/faro-experiments) with one example, two copy-and-fill templates, and notebook tests. They run on virtual-microscope-teaching (vmteach), a small pure-Python simulator of light-responsive cells behind the pymmcore-plus API, which replaces virtual-microscope as the base dependency. One code change under faro/: the simulated microscope now uses the real DMD class instead of an identity stub, so the calibration path is exercised on the virtual scope.

• examples/live_experiment.ipynb: a complete feedback experiment on the virtual microscope. The four objects, custom segmentator / feature extractor / stimulator, napari with stock napari-micromanager, DMD calibration, three phases, validate and load, staged start, continue, and a result figure showing every cell moving toward the light. Frames image phase contrast plus DAPI; nuclei are segmented on DAPI and grown to cell bodies; stimulation goes through the CyanStim channel, the only light path the teaching sim delivers the SLM pattern on. Two napari screenshots are taken by the notebook itself with nbscreenshot; committed with outputs.
• templates/live_experiment/ and templates/reanalysis/: each a uv project (pyproject.toml with the faro pin options) plus a notebook, plus a custom.py for the segmentator, feature extractor or stimulator that only one experiment needs, so the re-analysis notebook can import the same classes (a convention, not enforced). The live template runs end to end on the virtual microscope as shipped, including the DMD calibration cell; its TODO cells hold working defaults with the real-scope alternatives as comments. The re-analysis template only needs its paths cell filled in.
• templates/README.md: copy, uv sync, pin faro to a commit, update the pin, work against a local checkout, where experiment-specific code goes, all written for people new to uv.
• faro/microscope/simulation.py: UniMMCoreSimulation wraps the SLM in faro.core.dmd.DMD, uncalibrated until calibrate_dmd runs, like the real scopes. Microscope.calibrate_dmd accepts a Channel, a channel dict or a config name (the template already passed a name). The identity SimDMD stub moves to tests/fake_microscope.py, its only remaining user.
• tests/test_examples.py: executes the example and both templates in a real kernel on the virtual microscope (gui cells dropped, a parameters cell overridden with small frame counts, the re-analysis test generates its own source run), and checks that every README anchor the notebooks link to exists. New pytest marker examples; nbclient added to the test extras.
• tests/test_simulation_dmd.py: runs the real calibration routine against the teaching sim, whose projector can be misaligned via sim.slm_affine. A projector with 0.92 scale, 4° rotation and a 30 px shift is recovered to under 1 px over the field, and a mask pushed through the result lands on target.
• README: examples table, new "Writing your own components" and "Validation" sections, the three links into experiments/ retargeted, a fix for the Quickstart's non-existent MoveUp stimulator (and a calibrate_dmd line), extras table spelling, Installation now shows how to open the example, and the remaining em dashes are gone.
• experiments/ removed in its own commit (bad8434) so it can be dropped if wanted.
• pyproject.toml: virtual-microscope-teaching is a base dependency (git source, it is not on PyPI) so the example and the templates run right after uv sync. It is pure Python and brings numba, opencv-headless and matplotlib. The virtual-microscope extra is gone, as is imageio[ffmpeg], which nothing in faro imports.

Depends on

#26 (fix/omezarr-continue): the example's continue cell needs the OME-Zarr time-axis fix. Without it the continued frames are dropped and show up as storage errors in that cell (the committed outputs were generated with the fix applied). Merge #26 first.

Follow-ups (not in this PR)

• faro-experiments README still says to copy examples/experiment_template; it should point at templates/live_experiment. Superseded folders 01, 02, 03, 90 (and 99_demo_data unless kept as a try-out dataset) can go there.
• RunStatus from wait() reports 0 background errors even when the storage worker logged some, because analyzer errors are only folded in at finish_experiment. The example prints the status before that, so the count is misleading; worth fixing separately.
• DMD.calibrate leaves the core exposure at its 25 ms default; runs are unaffected since events carry their own exposure, but hand snaps after a calibration are dim.
• Two pre-existing test failures on this branch in a fresh environment (MDASequence compatibility, wait shift) are dependency drift, not this branch.
• The two window screenshots make the notebook about 2.3 MB.

https://claude.ai/code/session_01CSrNm4WLiNzDkiCi75vZfy