[tgockel]

A reply the provider ended with finish_reason: "content_filter" and no content read only as rig's "Response contained no message or tool call (empty)". RetryingModel passed rig's ResponseError through unchanged, and the retry line, model.attempt.error, model.retry.error, the failover's error, model.round.failed.error and run-new's [outrig] error: line are all made from it. RetryingModel::attempt now puts the finish reason the answered body names first in that message -- finish reason content_filter: Response contained ... -- first so the status line's 60 characters keep it.

The page and the screen also showed less than the events carry:

• render-session.py names an attempt's finish reason, says a failed one failed, and opens the error of an attempt, a retry and a move. A field a handler does not read is shown as a collapsed part named for it, so a field added later is not dropped unseen; a handler marks what it leaves off on purpose with Read.leave. Retries and moves now show their call and attempt ids, and a round its total and descendants.
• run-new's status line names the model a call moved from and why, until the model it moved to answers. An unfamiliar event's failure fields are taken out of its JSON before the 400-character cut and shown whole, so a hosted call's traceback is not cut. A child agent's failed round is a detail entry naming the child, and model.usage.refused is shown as an unfamiliar kind.
• Line mode prints the notes the screen puts in the conversation, through the same code, which adds a bindings process's exit cause.

The issue asked for a test that fails when a new failure field is not shown. In outrig-cli the events are #[non_exhaustive], so no test there can notice a new field, and there is none; on the page the fallback covers it instead. A failed attempt gets no screen line of its own: the retry, move or failed round that follows it carries the same text.

outrig run's copy of the retry layer never holds the body, so its text is unchanged; plan/next/outrig-run-names-the-finish-reason.md has the work.

Fixes #721

Summary

What does this change, and why?

Checklist

• cargo fmt --all -- --check passes
• cargo clippy --workspace --all-targets -- -D warnings is clean
• cargo test --workspace passes
• Docs in doc/ updated if behavior changed (mdbook build still clean)
• CHANGELOG updated for user-visible changes

Notes for reviewers

Anything worth calling out -- tricky bits, follow-ups, or intentionally out-of-scope items.