fix(gitea): revert cache workarounds, rely on CONFIG_FILE fix instead

Now that CONFIG_FILE is actually loaded by the daemon (see previous
commit), retest with the embedded cache server and default settings
instead of the dedicated cache-server/host.docker.internal workaround -
that whole setup was built to route around a config file that was
silently never being read in the first place.

gitea/runner-config.yaml

@@ -17,26 +17,12 @@ cache:
   dir: ""
   host: ""
   port: 0
-  # Both the shared network approaches we tried (a custom network name, then
-  # Docker's real "bridge" network) failed to make job containers reachable
-  # from/to this daemon's own embedded cache server - its host/port config
-  # is also unreliable in this act_runner version (confirmed by others: a
-  # fixed port still gets silently ignored in favor of a random one, see
-  # https://codeberg.org/forgejo/docs/issues/996). Using a dedicated
-  # `cache-server` process instead (see runner-cache-server in compose.yml),
-  # whose --host/--port CLI flags are documented and confirmed to actually
-  # work, reached via host.docker.internal + a published fixed port (proven
-  # reachable end-to-end from an isolated per-task-style container).
-  external_server: "http://host.docker.internal:8088/"
-  external_secret_file: "/data/cache_secret"
+  external_server: ""
 
 container:
   network: ""
   privileged: false
-  # --add-host is what makes host.docker.internal (used for
-  # cache.external_server above) resolve inside job containers on Linux -
-  # Docker only wires it up automatically on Docker Desktop (Mac/Windows).
-  options: "-v /var/run/docker.sock:/var/run/docker.sock --add-host=host.docker.internal:host-gateway"
+  options: "-v /var/run/docker.sock:/var/run/docker.sock"
   workdir_parent: ""
   valid_volumes: []
   docker_host: ""