fix(setup): fail early when a bind-mounted config file is missing

[badbread]

Hit in the wild while moving a long-running deployment back onto the stock compose.

The failure

docker-compose.yml bind-mounts two config files from the repo: caddy/Caddyfile and go2rtc/go2rtc.yaml. If one is missing, Docker doesn't say so. It silently creates a directory at that path, and the container then dies with:

error mounting ".../caddy/Caddyfile" to rootfs at "/etc/caddy/Caddyfile":
not a directory: Are you trying to mount a directory onto a file

That names the symptom, not the cause. Worse, it leaves the bogus directory behind, so the obvious next step (restore the file, retry) still fails until you rmdir it by hand.

Easy state to reach when a deployment keeps a partial copy of the repo instead of a full checkout — normal when running from prebuilt images and copying only what looks necessary.

Fix

A preflight in setup-env.sh. There's already precedent directly above it: the script prepares the backup directory specifically because "if Docker auto-creates the bind-mount dir it ends up root-owned and the api can't write." Same failure class, extended to the file mounts.

It handles both states:

• file missing → names the file and the git checkout to restore it
• path is a stray directory → tells you to rmdir it and restore, since that's what a user actually has by the time they investigate

Plus a troubleshooting entry for anyone who meets the raw Docker error before running setup.

Verification

Exercised all three cases against a scratch tree:

case	result
Caddyfile is a stray directory	✅ errors, tells you to rmdir
Caddyfile missing entirely	✅ errors, names the file
both files present	✅ passes

bash -n clean; confirmed it passes against a full checkout.

No Rust touched, so the Rust gate is unaffected.