Do not let the change listing kill the publish step

[slachiewicz]

The first run that got past authentication failed at the very last hurdle, on plexus-utils:

M  apidocs/org/codehaus/plexus/util/StringUtils.html
M  apidocs/org/codehaus/plexus/util/SweeperPool.html
...
##[error]Process completed with exit code 141

141 is SIGPIPE, and it is my bug, in this line:

git status --short | head -50

head exits after 50 lines. git status is then killed writing to a closed pipe. Under set -o pipefail that becomes the pipeline's exit status, and set -e makes it fatal.

Reproduced exactly:

$ bash -c 'set -euo pipefail; seq 1 500000 | head -50 >/dev/null'   # exit 141
$ bash -c 'set -euo pipefail; seq 1 500000 | sed -n "1,50p" >/dev/null'  # exit 0

It only bites when the output exceeds the pipe buffer, so this would have published a small site perfectly well and failed on a large one — plexus-utils regenerates thousands of apidocs files. A nastier failure mode than a consistent one.

The fix uses sed, which reads to end of input instead of exiting early, and says how many files were omitted from the listing.

The good news

Nothing was published, and everything before this worked. The step died before the commit, and the log shows the clone, the token authentication and the rsync all succeeded — the run had a fully populated gh-pages checkout with the new site staged. So the approach in #66 is sound; this is the last thing between it and a working publish.

Worth saying plainly: two failed runs, both mine, both found by actually running it rather than by reading it. The dry-run input has paid for itself twice now.