Template matching benchmarks all use the same hardcoded method

[sby1ce]

The macro that generates the benchmarks for template matching ignores the method parameter and uses hardcoded SQDIFF.

https://github.com/image-rs/imageproc/blob/ab9e19c241844996dfe0758a0e2f9e893b793142/src/template_matching.rs#L853-L866

Also cross correlation is not benchmarked at all.

Perhaps I'm misunderstanding how the macro works or what's the benchmarking process for this crate...