Don't error on broken pipe during a partial write to stdout

[str4d]

In #91 a user pointed out that a broken pipe during a partial write to stdout can occur as a normal matter of course (either due to ending a pseudo-seeking consumer like less early, or only needing a prefix of the decrypted file). We should not make this a hard error, and instead log a warning (if logging is enabled).

A broken pipe after no reading should IMO still be an error; if a user is explicitly trying to decrypt a file, but no downstream tool makes any attempt to use it, that seems like a bug in either the downstream tools or the user's command construction, and silently doing nothing would be terrible UX.