[Impeller] Change Path::CreatePolyline from tolerance to scale, and make it required

[dnfield]

Fixes flutter/flutter#121540
See also dnfield/flutter_svg#830 which has screenshots.

This patch mainly does two things:

• Makes sure that we pass a non-default tolerance value when creating a polyline for strokes. Previously, we calculated a tolerance but the method we called dropped the value on the floor (see first line of StrokePathGeometry::CreateSolidStrokeVertices). An "unused parameter" check wouldn't have caught this, since we use the parameter later in the method.
• Removes the stroke width from the calculation of the tolerance. This is error prone because a stroke_width of 0 means hairline for Flutter, but div by zero results in badness. I am not seeing any benefits when looking at this at various scales/high stroke widths.
• Changes the tolerance parameter on Path::CreatePolyilne to a scale parameter and removes the default. This should prevent this error from happening in the future.

Ideally we'd also have a test that CreateSolidStrokeVertices. That would require either friendship or making it public. I'm open to doing that if reviewers think it would be helpful. We do risk regression here if someone uses an incorrect value for the scale in the future.

[jonahwilliams]

LGTM

[chinmaygarde]

An "unused parameter" check wouldn't have caught this, since we use the parameter later in the method.

I think this makes sense. If only for TUs handled by the impeller_component GN template. We can fixup existing issues by either dropping the argument name or explicitly annotating it as being unused.

EDIT: Can we file a bug this?

[bdero]

LGTM. I have a questions about what should happen scale=0, but that concern is preexisting/not introduced by this PR.

[bdero]

IIRC, it's acceptable for tolerance to end up being NaN (when the max basis length is 0). Is that right?

[dnfield]

Right, curves ending reverting to lines.

[dnfield]

#38497 (comment) is where you discussed this before

[bdero]

Nit: Maybe mention what will happen when scale is 0 here. E.g. will curves revert to lines (seems like the most reasonable behavior)?

[dnfield]

Done