Ability to ignore unidcode decode errors.

I ran on a issue today where I as trying to read a File object from a malformed utf-8 string with an invalid character. I managed to fix my issue creating my own readlines() function with a try/catch block, but I don't think this solution is resistant to future updates from this libary since I cant' guarantee that the file cursor will move to the next line before this error occur.

def readlines(file):
  lines = []
  eof = False
  while not eof:
    try:
      line = file.readline()
      if line == '':
        eof = True
      else:
        lines.append(line)
    except UnicodeDecodeError as e:
      lines.append(e.object.decode('utf-8',errors='replace'))
  return lines

One simple solution to this would be adding a default error handling in line 68.

https://github.com/paramiko/paramiko/blob/cf7d49d66f3b1fbc8b0853518a54050182b3b5eb/paramiko/py3compat.py#L60-L71

changing to

return s.decode(encoding, errors='replace')

or even better giving the user the choice of how she/he would handle the errors via an optional argument.

Start in py3compat.py around lines 60–71, then reproduce the malformed UTF-8 File.readline case described in the issue. Review how decoding errors are currently surfaced and determine the intended replacement-versus-configurable behavior; done means the chosen handling is documented and the malformed-input behavior is covered by validation.