Introduce IFUNC facilities and port more optimizations onto aarch64

Enhancement

In #5248, TIFLASH_MULTIVERSIONED_VECTORIZATION is introduced to automatically generate IFUNC dispatched functions for TiFlash.

As an example, deserializeBinarySSE2 is updated to deserializeBinaryBlockImpl, where manually written inline asm for SSE2 optimization is replaced with tiflash_compiler_builtin_memcpy, which generates vectorized machine code of same quality with the original ones.
In the same time, a special version for x86-64 targets is provided, which utilizing ERMS technique for very long strings.

In #5251, more changes on top of #5248 are introduced, focusing on dispatching DMVersionFilterBlockInputStream.

As a clarification and a tip for future: clang 13 does not yet support auto vectorization utilizing SVE/SVE2 but clang 14 is ready for doing so. The code take SVE/SVE2 cases into consideration and it is tested to work on godbolt.

Start by reading #5248 and #5251, then trace the mentioned deserializeBinarySSE2/deserializeBinaryBlockImpl and DMVersionFilterBlockInputStream paths. Review how tiflash_compiler_builtin_memcpy and the clang 13/14 SVE/SVE2 behavior relate to the proposed dispatching. Done means introducing the IFUNC facilities and porting the identified optimizations to aarch64 while preserving the stated x86-64 behavior.