readpass_file: reject a passphrase file with a stray carriage return

[woahwhattheheck]

Disclosure, per this repository's AGENTS.md: I am an LLM (Claude), submitting on behalf of the account owner. I am available to discuss this change and to revise it in response to review feedback.

Fixes #431.

readpass_file() ends with

	/* Truncate at any newline character. */
	passbuf[strcspn(passbuf, "\r\n")] = '\0';

fgets() stops after a \n, and the fgetc(f) != EOF check above rejects
anything following it, so a \n in the buffer is always the last character.
A \r is not: fgets() reads past it, so a \r anywhere in the line reaches
strcspn() and silently cuts the passphrase off there.

--passphrase file:pw with pw containing hunter2\rextra therefore uses
hunter2. It round-trips, because encryption truncates the same way, so
nothing shows until the passphrase is supplied by another route. The header
in libcperciva/util/readpass.h already says a file like that should be
rejected:

Print an error and fail if the file is 2048 characters or more, or if it
contains any newline \n or \r\n characters other than at the end of the
file.

The change

Strip a trailing \n or \r\n explicitly, then reject the file if any \r
or \n is left. I did not make the leftover character part of the
passphrase: both that and the current truncation change the passphrase an
existing file yields, and neither is safe to choose on the user's behalf
without telling them. No documentation change — this makes the code match
what readpass.h already describes.

Files whose only newline is a \n or a trailing \r\n are unaffected, so
ordinary Unix and Windows passphrase files keep working.

Test

tests/08-passphrase-file.sh gains four checks rather than a new test file:

• a passphrase file containing hunter2\rextra must be rejected, with the
  error message, and must not create an output file. The passphrase begins
  with the correct one, so before this change that decryption succeeds and
  the check fails — it is a real regression test, not a restatement.
• a passphrase file with CRLF line endings must still decrypt the reference
  file correctly.

Note

This is libcperciva code, so it also lives in tarsnap and kivaloo; I have
only touched the copy in this repository.