Fix IO.read/2 docs: integer count is in characters, not bytes

[nseaSeb]

IO.read/2 is the Unicode-safe reader and delegates an integer count to :io.get_chars/3, which counts characters, not bytes. The docs were copied from IO.binread/2 (the byte-oriented, Unicode-unsafe reader), so they incorrectly described the integer argument as a number of bytes and carried binread/2's "preferred mode for reading non-textual inputs" note. This also contradicted read/2's own return description, which already documents the result as "the output characters".

Assisted-by: Claude Code:claude-opus-4-8

[AlexGx]

IMO its a not fully correct change. Characters is too broad and imprecise.

If the device is in Unicode mode, the count is number of Unicode code points. Otherwise, the count is the number of bytes.

Quick proof

echo "José" > test.txt

iex(1)> {:ok, file} = File.open("test.txt")
{:ok, #PID<0.109.0>}
iex(2)> IO.read(file, 4)
"JosÃ"
iex(3)> {:ok, file2} = File.open("test.txt", [:utf8])
{:ok, #PID<0.110.0>}
iex(4)> IO.read(file2, 4)
"José"

Also Id suggest to update returns section to smth like:

- `data` - the data read from the device as a binary or charlist,
    depending on the device mode

[josevalim]

Thanks, I pushed a clearer description!