This capability is required for read pipelining. We will only read in as
many records as will fit in the read buffer (and the network can provide
in one go). The bigger the buffer the more records we can process in
parallel.
Reviewed-by: NTim Hudson <tjh@openssl.org>