This commit adds expect_finite_number() and finite_f32() to parsers.rs.

Part of issue #344 is that we need to validate parsed numbers, as
cssparser can give us back -inf or inf when it overflows.  These
values ripple down the code and end up creating a NaN inside a
cairo::Matrix, which makes it invalid.

This commit introduces a CssParserExt trait to have an
expect_finite_number() method for cssparser::Parser, and the
corresponding, low-level finite_f32() validation function.

Those functions return a Result<f32, ValueErrorKind>.  This commit
makes the following return a ValueErrorKind for the error case as
well:

* impl Parse for f64
* parsers::number()
* parsers::number_optional_number()
* parsers:;integer()
* parsers::integer_optional_integer()

This has implications in all the callers, since they used to expect a
ParseError, but must now expect ValueErrorKind.

The biggest change is in transform.rs.  For simplicity, the individual
parsers for transform functions now have two sections:  the first
extracts f32 from cssparser::Parser, and the second validates those
numbers and computes the final result --- instead of doing everything
at the same time.

#344