Fix newline for empty prompts

[squell]

Closes #1648

Note: this does mean that in irregular exits (e.g. due to ?, or a panic) of prompt_password, the new line would not be printed. But the only irregular that is there is the SIGTSTP/SIGTTIN/SIGTTOU, and actually with this patch the behaviour of those seems closer to ogsudo as well (e.g. ^Z will cause a newline to be printed anyway)

In particular, any I/O errors in read_unbuffered are handled by prompt_password and not propagated.

[bjorn3]

The write could be moved above the take_first_pending() if we do want a newline on signals, but the current location is fine too.

Would a test make sense?

[squell]

The write could be moved above the take_first_pending() if we do want a newline on signals, but the current location is fine too.

I'll experiment with that.

Would a test make sense?

We can take the unit test from #1654, those seem small enough. Although it also feels like compliance-test territory.