feat: give the error page a way onward

[ashleycaselli]

Closes #616.

ErrorPage said what went wrong and nothing else — no link home, no way back to what the user was doing, no hint of who could act on it. Compare NanopubNotFoundPage, which explains the situation, offers a retry and a link home, and answers with a fitting HTTP status.

The shape of the error, not just its details

A new ErrorPage.Kind, set through a kind page parameter next to the existing message, says whose problem it is. It decides what the page tells the user to do, whether it offers a way to report the problem, and the status it is answered with:

Kind	What the page says	Status
REQUEST	The address is wrong; you — or whoever wrote the link — can correct it	400
NOT_FOUND	There is no page at this address (which it spells out)	404
CONTENT	The published nanopublication is at fault; only its author can put it right, by publishing a corrected version	422
MALFUNCTION	This one is on Nanodash, and here is how to report it	500

configureResponse() answers with the matching status the way NanopubNotFoundPage does, so non-browser clients can tell these apart too. Without a kind, the page falls back to the status the servlet container settled on — that is how web.xml's /error/404 and /error/500 arrive — and to MALFUNCTION otherwise, since an error nobody foresaw usually is one.

A way onward

• A link home on every error page.
• A link back to where the user was, built from the Referer and accepted only for this instance's own pages — an off-site referrer is not where they were working — and never for an error page itself, which would only loop.
• For the errors that are ours, a pre-filled issue on the tracker carrying what the user saw, the failing address, where they came from, and the Nanodash version, so a malfunction doesn't dead-end at a stack trace in the server log.

Callers say what they ran into

• NanopubNotFoundPage.forwardFor() marks a malformed identifier as the asking user's to correct.
• GrlcQuery classifies a failed load — the asking user's when the ID holds no artifact code at all, so nothing could have been fetched for it, and the query author's otherwise — and passes it through QueryLoadException.getKind() to QueryPage.
• The "which only a corrected version of it can fix" sentence moves out of the SPARQL message and into the page, where it now covers every error of that kind rather than just that one.

Tests

11 tests in ErrorPageTest cover the per-kind wording and status, the report link's presence and contents, the back-link's accept and reject cases, and the container-status takeover. Kind assertions were added to GrlcQueryTest, QueryPageTest, and NanopubNotFoundPageTest. Full suite green: 1134 tests, 0 failures.

🤖 Generated with Claude Code

[github-actions]

🎉 This PR is included in version 5.12.0 🎉

The release is available on:

• nanodash-5.12.0
• GitHub release

Your semantic-release bot 📦🚀