feat(backend): per-Reader userscript credential with UI install and rotation (#24)

Each Reader's userscript credential is derived from TOKEN_KEY, their
Discord id and a token epoch (HMAC-SHA256, hex); only its SHA-256 sits in
readers.token_sha256, so install URLs survive restarts while a database
leak yields nothing but hashes. One credential authenticates the script
download path and the API bearer header.

- internal/token: derivation + hashing; migration 0006 adds token_epoch
- seed refreshes the owner's epoch-0 hash only before first rotation
- httpmw.Auth resolves the acting Reader from the credential hash and
  stashes it in the request context; the retired API_TOKEN resolves to
  the owner until API_TOKEN_GRACE_UNTIL, logged per use, on both the
  bearer and script-download paths
- userscript handler renders the bindmounted file with the resolved
  Reader's credential substituted for __API_TOKEN__; a legacy-path
  request during grace serves the derived credential, so devices
  self-migrate on their next update poll
- web UI: Userscripts panel with session-gated install endpoints that
  render the script directly (credential never in markup, address bar
  or a redirect) and confirm-gated rotation; atomic epoch bump + hash
  rewrite in the store
- both userscripts carry __API_TOKEN__ placeholders; the committed
  global-token literal is removed (rotating at deploy retires it for
  real — it survives in git history)
- env: TOKEN_KEY required, API_TOKEN/API_TOKEN_GRACE_UNTIL retire the
  legacy credential; docs and compose updated

REDEPLOY.md

@@ -224,6 +224,8 @@ Same four API checks as `DEPLOY.md` §3, plus the web UI. Set the host names onc
 ```bash
 API=https://bookmark-api.violetcrown.my.id
 WEB=https://bookmark.violetcrown.my.id
+# During the grace window the retired global credential still resolves to the
+# owner; afterwards it is 401 like any other wrong credential.
 TOKEN=$(grep -E '^API_TOKEN=' .env | cut -d= -f2)
 
 curl -s $API/healthz                                            # -> ok