[tools] Second take to fix bloat check (#26544)

Apparently, in the previous PR I replaced the deprecated
DataFrame.append with pandas.concat, but I didn't realize
that our scripts add extra attributes to the data frame,
which are not copied during the concatenation.

This time, I verified the script locally using
--github-comment and --github-dryrun-comment.

Also, do not fetch submodules in the bloat check job as
it takes most of the job's execution time.

.github/workflows/bloat_check.yaml

@@ -41,7 +41,6 @@ jobs:
  with:
  action: actions/checkout@v3
  with: |
- submodules: true
  token: ${{ github.token }}
  attempt_limit: 3
  attempt_delay: 2000

scripts/tools/memory/gh_report.py

@@ -375,8 +375,10 @@ def merge(df: pd.DataFrame, comment) -> pd.DataFrame:
  cols, rows = memdf.util.markdown.read_hierified(body)
  break
  logging.debug('REC: read %d rows', len(rows))
+ attrs = df.attrs
  df = pd.concat([df, pd.DataFrame(data=rows, columns=cols).astype(df.dtypes)],
  ignore_index=True)
+ df.attrs = attrs
  return df.sort_values(
  by=['platform', 'target', 'config', 'section']).drop_duplicates()