Fixes #12165

Summary

When min, max, and stepSize are set, linear tick generation used a spacing-based tolerance (spacing / 1000) to detect whole step counts.
With larger step sizes, that tolerance can be too loose and misclassify non-divisible ranges as divisible, producing evenly distributed ticks instead of stepSize increments.

This change switches the whole-step check to a machine-precision-based epsilon derived from stepCount.

Behavior after fix

For non-divisible ranges, ticks follow stepSize multiples and keep max as the final shorter interval:

• min: 0, max: 3333, stepSize: 500
• ticks: [0, 500, 1000, 1500, 2000, 2500, 3000, 3333]

Floating-point divisible ranges still behave correctly:

• min: 0, max: 0.3, stepSize: 0.1

Tests

Added regression tests in test/specs/scale.linear.tests.js for:

1. Large non-divisible range (0..3333, stepSize: 500)
2. Floating-point divisible range (0..0.3, stepSize: 0.1)

Validated with:

• pnpm run lint-js
• pnpm run test-ci-karma scale.linear.tests

References

• Issue repro: Inconsistent step size behaviour when step size is not a divisor of (max - min) #12165
• CodePen (reported case): https://codepen.io/Pia-Gerhofer/pen/LENMKOO