You can set readline's editing mode to vi so that anything that uses readline will then have vi-style bindings. Very useful, particularly when combined with rlwrap for those few programs that don't have readline support.
Zsh doesn't use readline (it has it's own line editor) but it can be set to vi-style input too. (Actually, at least on my system, bash doesn't seem to be linked with libreadline.. I'm not sure if it is actually using readline or not)
It looks like bash does use readline, though it's not dynamically linked with it (at least on my system). Kind of seems like bash and readline are intertwined more than I thought; the git repo for bash has a distribution of readline included in it.
That makes sense to me. The shell is fundamental to the system, and is so heavily intertwined with readline that it is very sensitive to ABI/API changes. Consider: if you upgrade your readline without upgrading bash, or vice versa, you lose the ability to fix the problem even in single user mode. Having Bash control its own readline usage seems like the safest option, though conceptually a bit more complex.
Zsh doesn't use readline (it has it's own line editor) but it can be set to vi-style input too. (Actually, at least on my system, bash doesn't seem to be linked with libreadline.. I'm not sure if it is actually using readline or not)