fix: guard TemperatureCache.complement against a zero temperature range

[yuksblog]

Fixes #28

The bug

A seed whose HCT sits at a tone extreme (#FFFFFF, #000000, #010101) collapses the gamut to a single point, so every hue in the 361-entry sweep resolves to the same color and the warmest and coldest entries are identical. complement divided by that zero range and raised ZeroDivisionError:

from materialyoucolor.hct.hct import Hct
from materialyoucolor.scheme.scheme_fidelity import SchemeFidelity

SchemeFidelity(Hct.from_int(0xFFFFFFFF), False, 0.0, spec_version="2021")
# ZeroDivisionError: float division by zero

#FFFFFF is an ordinary seed for a user to pick, so this is reachable from normal input, via get_tertiary_palette in any scheme that needs the complement.

The fix

Return the input when the range is zero: every candidate has the same temperature, so none is a better complement than another. This mirrors the zero-range case that relative_temperature already handles a few lines below.

Verification

Non-degenerate seeds are bit-for-bit unchanged. I compared every color of all 9 dynamic schemes across 5 seeds x light/dark x 4 contrast levels x both spec versions (42,480 values) before and after the change: identical.

The regression check went into tests/test_all.py, since that is what CI already runs and it fails the build on a non-zero exit — without the fix in the first commit, the second commit makes the wheel build go red. It sits ahead of the early exit on Windows so it runs on every platform. Happy to drop that commit if you would rather keep test_all.py purely visual.

[T-Dynamos]

LGTM - Thanks.