Refactors fromString implementation

[chrvadala]

Scope

This PR refactors fromString(matrix) function with the following benefits:

• removes the used RegExp that has the following risk https://codeql.github.com/codeql-query-help/javascript/js-polynomial-redos/
• supports optional measure unit (e.g. 100px). This leverages on parseFloat() that ignores them.
• validates provided numbers, throwing exception in case of !Number.isFinite() (e.g. two)

⚠️ breaking changes

Previous implementation where not fully validating number, providing in some case Number.NaN as output. This new version strictly validate number and may break some part of your code that were silently failing.

This table shows some examples:

parsed string	Behaviour versions <=2.17	Behaviour now
matrix(ee,ee,ee,ee,ee,ee)	{ a: NaN, b: NaN, c: NaN, d: NaN, e: NaN, f: NaN }	Error("'matrix(ee,ee,ee,ee,ee,ee)' is not a matrix")
matrix(1px,2px,3px,4px,5px,6px)	Error("'matrix(1px,2px,3px,4px,5px,6px)' is not a matrix")	{ a: 1, b: 2, c: 3, d: 4, e: 5, f: 6 }

🙅 backward compatibility

In a case that you want still to use the previous implementation, you can replace fromString(matrix) with fromStringLegacy(matrix). At the same time, open an issue in order to show your use case, because future implementations will remove the legacy method.

[coveralls]

coverage: 100.0%. remained the same
when pulling 85a094e on fromStringV2
into 31aba56 on main.