ci/cache: Keep collection in preflight

Group cache collection separately in the Actions log and avoid repeating
it after updating the CI profile.

.gitea/workflows/ci.yaml

@@ -37,9 +37,9 @@ jobs:
         env:
           HARMONIA_SSH_KEY: ${{ secrets.HARMONIA_SSH_KEY }}
         run: |
-          # Make room before pushing so a full cache cannot prevent the GC at
-          # the end of a successful run from ever being reached.
+          echo "::group::Collect cache garbage"
           ci/push-to-cache.sh --gc
+          echo "::endgroup::"
 
           nix eval --json --apply "builtins.attrNames" .#ci.x86_64-linux | jq -cr '.[]' | while read job; do
             echo "::group::Build $job"

ci/push-to-cache.sh

@@ -39,6 +39,4 @@ nix copy --no-check-sigs --to "$STORE_URI" "$path"
 if [ -n "$UPDATE_PROFILE" ]; then
   echo "Updating profile..."
   remote_cmd nix-env -p "$REMOTE_STORE"/nix/var/nix/profiles/nixfiles --set "$path"
-
-  collect_garbage
 fi

docs/deployment.md

@@ -227,10 +227,9 @@ workflow deletes cache-profile generations older than its retention period, then
 unreachable paths so a full cache cannot prevent collection from being reached.
 
 It then builds `.#ciDrv.x86_64-linux`, a `linkFarm` of all CI attributes, and pushes it with
-`UPDATE_PROFILE=1`. That updates the `nixfiles` profile on the cache box and collects old paths
-according to the workflow's retention setting. The SSH store uses `/var/lib/harmonia`,
-`HARMONIA_SSH_KEY`, and pinned `ci/known_hosts`; clients use `https://nix-cache.nul.ie` through
-`lib.my.c.nix.cache`.
+`UPDATE_PROFILE=1`. That updates the `nixfiles` profile on the cache box. The SSH store uses
+`/var/lib/harmonia`, `HARMONIA_SSH_KEY`, and pinned `ci/known_hosts`; clients use
+`https://nix-cache.nul.ie` through `lib.my.c.nix.cache`.
 
 ### `installer.yaml`