Flag the unverified spectrum wavelength axis; keep the fixture generator

The linear 380-1050 nm axis gives spectrum chromaticities 0.09-0.13 (xy)
away from the instrument's own x, y, so TM-30 and spd.csv inherit an axis
error until it is confirmed. tools/make_fixtures.py regenerates the golden
data from the python-final driver. The release profile moves to the
workspace that builds the binary.

PROTOCOL.md

@@ -190,6 +190,19 @@ The 350 floats cover **380 nm to 1050 nm** at ~1.92 nm steps
 ((1050-380) / (350-1) = 1.917 nm/step).  Values are absolute spectral
 irradiance; the vendor software normalizes by dividing by the peak value.
 
+> **Unverified: the wavelength axis.** The linear 380-1050 nm mapping above
+> is an assumption. Integrating the published lamp spectra with the CIE 1931
+> observer on this axis gives chromaticities 0.09-0.13 (xy) away from the
+> x, y the instrument itself reports in the same block (offsets 52/56); for
+> L0011 the spectrum says x,y = 0.358, 0.295 against the instrument's 0.380,
+> 0.382. No other linear range fits all lamps (380-780 nm comes closer but
+> moves the blue LED peak to ~419 nm). The 350 points are probably on a
+> non-linear (spectrometer pixel) axis. Everything computed from the
+> spectrum (TM-30 Rf/Rg, the spd.csv files) inherits this until the axis is
+> confirmed, e.g. against a spectrum exported by the vendor software with
+> its wavelengths, or by fitting the axis so the spectrum reproduces the
+> instrument's own XYZ.
+
 ### Unmapped Regions
 
 Offsets 176-223, 256-271, and 1832-3903 contain additional data not yet