Propagate metadata upload failures through setup

[youseihuayu-wonderful]

Closes #12804. Related to #12467.

Problem

The shared metadata uploader ignored HTTP errors: an HTTP 500 could still end with exit 0 and "Done". Bootstrap and the installer also use this script.

Fix

• Check HTTP and curl status; report the failed block, status and response body.
• Attempt all standard blocks and return non-zero if any failed.
• Finish remaining setup/persona configuration and security lockdown before returning failure. Preserve dev's intentionally insecure mode and installer step 8c.
• Keep 10156 fast redeploy maven #12467 from reaching Solr updates or "Done" after a metadata failure.

Verification

• Shellspec: 44 examples, 0 failures, 1 existing skip; stricter POST/content-type/file checks and lockdown/status regressions.
• Live Docker/Dataverse: valid TSV persisted a new field; HTTP 500 identified geospatial TSV line 42.
• Failed dev/demo bootstrap still created admin/root and completed persona configuration. Demo admin APIs remained blocked; dev exported its token.
• Actual packaged installer post-deployment stage completed lockdown and remote Solr configuration, then exited 0/1 as appropriate. Full host provisioning was not tested.
• 10156 fast redeploy maven #12467 exited 1 before Solr updates or "Done".

Guide/release note updated. Successful block uploads are not rolled back. Implementation and verification were AI-assisted.

[ErykKul]

Thanks for picking this up! The uploader change looks good. The main issue is the early exit: a failed block now skips the rest of setup, including the admin API lockdown, and leaves an instance that can't be re-bootstrapped. The inline comments on setup-all.sh and install.py suggest letting setup finish and exiting non-zero at the end, plus a test for that.

[youseihuayu-wonderful]

@ErykKul Thanks — addressed in 17aa7aa.

Setup now retains the metadata failure, finishes remaining configuration and security lockdown, then exits non-zero. Dev stays intentionally insecure; demo completes its lockdown; the installer still runs step 8c; bootstrap preserves env export.

Added lockdown/status regressions and stricter curl request checks: 44 examples, 0 failures, 1 existing skip. Live Docker checks confirmed admin/root creation, demo/installer API blocking, and installer exit 0/1. #12467 still stops before Solr and "Done" on HTTP 500. Guide/release note corrected.

Installer verification covers its actual packaged post-deployment stage, not full host provisioning. Ready for re-review.

[youseihuayu-wonderful]

@ErykKul
Hi Eryk, I’ve updated #12805 (#12805) to address the review comments. Could you take another look? The JSF job failed while pulling the Keycloak image from quay.io with a 504 error—could you rerun that job?

[ErykKul]

Looks good to me, approved. QA'd with fresh unstable images and this PR's scripts, using a TSV with an invalid displayOrder:

• dev bootstrap, valid TSVs: exit 0, everything configured
• dev bootstrap, broken TSV: the error shows the block, the HTTP 500 and the TSV line; the rest of setup completes; exit 1
• demo bootstrap, broken TSV: lockdown still applied (admin API blocked, burrito revoked); exit 1

#12467's update-metadata.sh: stops before Solr and "Done" on failure, works normally otherwise
Not tested: the host installer.