Record reports after the upload, not before - #347

all-reports.py wrote a report's row into [history] as soon as it had
written the html to the workspace, and mailed everyone about it in the
same run. The files only reach the web server at the end of the stage,
after all-plots.py and two dozen report.py invocations, so anything
failing in between left a report that the database says exists, that
nobody can read, and that the rerun skips because it is recorded. The
mail linking to it had already gone out.

Before the table existed the record was the published 00_history.html,
which only existed if the upload had happened, so a lost upload healed
itself on the next run; moving the record into the database made it
permanent.

The reports and the queued emails now go into pending-reports.json, and
publish-reports.py records and sends them once the files are up. The
stage uploads history/ right after all-plots.py and keeps a second
transfer for overview*.html at the end.

| scenario                            | now                          |
| ----------------------------------- | ---------------------------- |
| generate, publish, rerun            | skipped                      |
| stage fails before the upload       | regenerated next run         |
| uploaded but not recorded           | copied back from the index   |
| branch named twice in one run       | one report                   |

Assisted-by: Claude Opus 5