Optimized `Lt128` operator for RVV

[lsrcz]

This pull request adds an optimized implementation of the Lt128 operator for RVV targets. The new implementation is synthesized using a program synthesizer.

The main computations use LMUL 1/8, which is usually more efficient than vector groups (LMUL > 1) and can outperform full vector registers (LMUL = 1) on some microarchitectures.

[lsrcz]

The compilation result: https://lt128.godbolt.org/z/xEK6v4f6f.