fix(pg-protocol): keep the bytes of binary columns instead of decoding them as text

[nigrosimone]

Fixes #3495. With binary: true the parser decodes every DataRow field as utf8 text, and Result turns the string back into a Buffer before the type parser runs. Any byte utf8 cannot carry is replaced on the way, so SELECT 1000::int4 comes back as 1007 and 200 as 239.

The row description says which columns are binary. The parser now remembers that and hands those fields over as bytes, the text columns as strings like before, so a query without binary columns pays one null check per row.

Two tests: the parser one feeds a binary row description and a row with the bytes of 1000, the pg one runs a binary query with an int4, a float8 and a non-ascii text. Both fail on master and pass here.

Built on #3776, which makes BufferReader.bytes return a copy; the first commit is that one.

[charmander]

#3734

[nigrosimone]

Oook 😅