mwish

@joellubi For arrow C++ usecase, encode is implemented like this patch now, decode, however is implemented by batch: ```c++ int DecodeRaw(uint8_t* out_buffer, int max_values) { const int values_to_decode = std::min(num_values_,...

> but there is a slight difference between how Go and cpp implementations batch reads. Yes, but this doesn't change the syntax for "streaming" read from values, right? I think...

ByteStreamSplit Part LGTM > Overall, the batched approach improves performance slightly across the board for decoding. This is most likely because an intermediary buffer is no longer needed with this...

I'll be out until 9.21, sorry for delaying. I think kou and pitrou can well review this