`Reader.read_row` is an alternative to `Reader.next_row`.

* `Reader.next_row` is convenient when the caller doesn't want to manage
  their own buffer, and doesn't mind having extended borrows on `Reader`
  to accommodate `Row` data.
* OTOH the new `Reader.read_row` avoids an extra copy in some scanarios
  which may lead to better runtime performance - see the benchmark
  results below.

The commit results in the followin performance gains seen in the
recently introduced `row-by-row/128x128-4k-idat` benchmark:

time:   [-16.414% -16.196% -15.969%] (p = 0.00 < 0.05)
time:   [-15.570% -15.218% -14.856%] (p = 0.00 < 0.05)
time:   [-16.101% -15.864% -15.629%] (p = 0.00 < 0.05)