[libc++abi] Improve performance of __dynamic_cast

The original `__dynamic_cast` implementation does not use the ABI-provided `src2dst_offset` parameter which helps improve performance on the hot paths. This patch improves the performance on the hot paths in `__dynamic_cast` by leveraging hints provided by the `src2dst_offset` parameter. This patch also includes a performance benchmark suite for the `__dynamic_cast` implementation.

Reviewed By: philnik, ldionne, #libc, #libc_abi, avogelsgesang

Spies: mikhail.ramalho, avogelsgesang, xingxue, libcxx-commits

Differential Revision: https://reviews.llvm.org/D138005