[PierreRaybaut]

Fixes #74

Several problems reported by DataLab 1.3.0 users about the blackbody signal, curve fitting and synthetic images.

Bug fixes

• Automatic Planckian fit: on data in physical units (e.g. a blackbody curve with wavelengths in meters), planckian_fit returned its initial guess unchanged. The fitted peak came out exactly exp(5) - 1 (about 147) times too low, with no warning. The initial guess now matches the data peak, and the fit runs on normalized data, so it converges whatever the units. With the default blackbody signal, the fit gives back T = 293 K and an emissivity of 1, to better than 1e-6 relative error.
• Exponential fit: the rate b was bounded to [-10, 10] whatever the X unit, so a 1 ms time constant with X in seconds could not be fitted. The bound now scales with 1/Δx, and evaluate clips the exponent b·x instead of b itself.

New features

• FitComputer.FORMULA and get_formula() give the rich-text formula of every fit model (single source of truth for SigimaX, DataLab and DataLab-Web).
• PlanckParam now has default axis titles and units: wavelength in m, spectral radiance in W·sr⁻¹·m⁻³. Its docstring had the wrong unit.
• Gaussian, 2D ramp, sinusoidal grating, ring, Siemens star and 2D sinc images now derive from DomainImageParam. Its "Physical coordinates" option (enabled by default) uses the [xmin, xmax] × [ymin, ymax] domain as the image coordinates. Parameters serialized without this option (creation metadata, workspaces, history) are deserialized with it disabled, so recreating an older image keeps its pixel coordinates.

Not changed

The 4-parameter Planck model and the fit_params schema are unchanged. The model is still over-parameterized (#43 stays open); the docstring now explains how to derive peak wavelength, temperature and emissivity from the fitted parameters.

Companion PRs (merge order: Sigima → SigimaX → DataLab)

• SigimaX: Show the model formula in interactive fit dialogs SigimaX#10
• DataLab: Fix Planckian fit, Planck axes and synthetic image coordinates DataLab#401

Validation

• Full test suite: 2191 passed, 34 skipped.
• New regression tests cover:
  • Planckian fit over several temperatures, with X in m, µm and nm, with noise and a baseline;
  • exponential fit with a 1 ms time constant;
  • a FORMULA for every fit type;
  • physical coordinates for each of the 6 image types, with the option on and off, plus legacy deserialization;
  • Planck axis labels.
• Ruff and Pylint (10/10) pass on the changed modules.