expose read_buffer_size in parts

[rizo]

When dealing with slightly large files (>1MB), the reader loop becomes unnecessary slow. Increasing the buffer size from the default 10 to something larger substantially speeds up processing.

I understand the non-blocking API can be used to provide a custom buffer size, but I think it should be possible to do this for the blocking version too when the expected inputs safely fit in memory.

Also, any reason why the read_buffer_size values are different for parts and reader? Why not just have one default?

[rizo]

I kept the default value to 10, but would be happy to make it just ?read_buffer_size so there's only one default.