perf: optimize vector marshal/unmarshal for float32/float64/int32/int64 (Throughput: 75 MiB/s → 1.9 GiB/s (marshal), 106 MiB/s → 4.6 GiB/s (unmarshal) )

perf: optimize vector marshal/unmarshal for float32/float64/int32/int64 (Throughput: 75 MiB/s → 1.9 GiB/s (marshal), 106 MiB/s → 4.6 GiB/s (unmarshal) ) #2284