ssh's password prompt doesn't use any special terminal-input library like
readline. The line-editing features are just the baseline POSIX TTY line-editing features.
So you have a POSIX TTY in "cooked" mode (not raw), aka canonical mode, and the only line editing that's available is what's provided by the kernel. See
stty(1), and notice that
kill = ^U. This is also where the backspace character is defined (
erase = ^?). Word-erase (
^W) is convenient when you're not typing blind.
lnext = ^V means you can type control-v then anything (including control-c) to get a literal control-c.
To debug what you were trying to do blindly, run
cat > /dev/null in your terminal. Type stuff, then see what works and what doesn't to edit it.
readline (used by
bash) reads raw character and does the line-editing in user-space. Its default bindings are compatible with the default TTY control characters, though, for the subset of editing features that they both provide.
readline goes way beyond the simple line editing of a plain TTY. (e.g. a TTY can only delete characters at the end of the line, so there's no
delete or left/right arrow)
bash runs a command in the foreground, it puts the TTY into canonical mode first (because that's the default). So running
stty -a (with no redirection) will always see its own terminal in canonical mode. But if you redirect input from some other TTY that has
bash running on it, you can see what terminal settings bash + readline applied. e.g.
stty -a < /dev/pts/12 shows
-icanon for raw mode because I have a
bash running on that terminal. (I switched to another tab and ran
tty, then used that device file path from the first terminal). If I ran
cat in that other terminal, I'd see
icanon for canonical mode.
Related: The TTY demystified