Support for Arm Scalable Vector Extension (SVE/SVE2)

Currently, the performance of simdjson on Arm-based platforms using the NEON implementation path is significantly lower than on x86 platforms. In our benchmarks, the NEON performance reaches only about 60% of what is achieved on x86-64.

I would like to propose the introduction of an SVE/SVE2 implementation path. SVE's predication, variable vector length support, and specialized instructions could potentially close the performance gap.

I am interested in contributing to the implementation and testing of the SVE implementation path. I would appreciate guidance from the maintainers on the preferred integration strategy within the existing dispatch framework.