Simplify workflow for deploying new HTML pages

Cloning the entire repo is very slow, due to the gh-pages branch
containing nearly 2GiB of static html files. The current version of the
deploy.yml workflow uses actions/cache@v4 to cache a copy of the full
repo which can be retrieved much faster than doing a clone. This makes
the workflow faster, but is complex to understand. It also doesn't help
if no changes have been pushed to the master branch for more than a
week, because the cache expires and so a full clone is needed anyway.

This change attempts to remove the cache so that the workflow is simpler
to understand and more robust.

To avoid cloning the full repo, the initial clone uses the magic number
fetch-depth:2147483647 i.e. INT_MAX. According to the documentation at
https://git-scm.com/docs/shallow this means infinite depth. However,
unlike fetch-depth:0 it implies the --single-branch option for the
clone, so only the master branch is fetched and not the huge gh-pages
branch.

The gh-pages branch is still needed in order to push to that branch, but
we clone that in a separate step (into a sub-directory) using
fetch-depth:1 to avoid fetching the full history. That step doesn't need
to change, as it wasn't using the cache anyway.

Author: jwakely

.github/workflows/deploy.yml

@@ -12,40 +12,13 @@ jobs:
     runs-on: ubuntu-latest
 
     steps:
-    - name: Use cached git repo
-      # Try to use a cached copy of the repository to avoid cloning it again,
-      # as cloning the full history (needed for timestamps) takes two minutes.
-      id: cache-repo
-      uses: actions/cache@v4
-      with:
-        path: |
-          .git
-          xml
-          meta-data
-          src
-        key: cached-repo
-
-    - if: ${{ steps.cache-repo.outputs.cache-hit == 'true' }}
-      # Using cached git repo, so sync with the latest changes.
-      name: Update cached repo
-      run: |
-        # Make sure we are in a clean worktree on the right branch.
-        git reset --hard HEAD
-        git checkout master
-        # Fetch changes since the cache was saved.
-        git fetch
-        # Update timestamp of this file so only issue files that get
-        # changed by the next command will be newer than it.
-        touch meta-data/dates
-        # Do a hard reset instead of merge, just in case there has been
-        # a force push since the cache was last updated.
-        git reset --hard @{u}
-
-    - if: ${{ steps.cache-repo.outputs.cache-hit != 'true' }}
-      # Not using cached git repo, so clone it (with full history).
+    - name: Clone XML sources
       uses: actions/checkout@v5
       with:
-        fetch-depth: 0
+        ref: ${{ env.BRANCH }}
+        # Using fetch-depth implies --single-branch so we don't fetch
+        # the gh-pages branch (which avoids fetching 99% of the objects).
+        fetch-depth: 2147483647
 
     - name: Update issue timestamps
       run: make dates
@@ -56,8 +29,7 @@ jobs:
     - name: Generate HTML lists
       run: make lists
 
-    # The cached repo might have a stale API token that can't push changes.
-    # Clone a fresh repo that has valid credentials to push to gh-pages.
+    # Clone the gh-pages branch into a sub-directory.
     # The default fetch-depth:1 is OK for this step.
     - name: Clone repo again in order to push
       uses: actions/checkout@v5