fix(web): show a Decision's places by their marks and a hidden place as hidden

[zfy0701]

Summary

Three rough edges on the Decision page's places card (#2561), seen on a real Decision in the test environment:

• Place marks. A channel gate showed a hash glyph beside a label that already starts with # (# #bot-test · …), and every other kind a generic glyph. The editor now passes a markFor beside hrefFor, so each tab and each row's source shows the gate's integration platform, the shared bot's platform, the code host (GitHub, GitLab, Gitea), or the agent's own icon for model selection and agent tools. The kind glyph remains the fallback while console data has not loaded.
• A place the viewer cannot read is not a failure. The conversation evaluations list answers 404 when the conversation's audience refuses the caller (for example, a channel whose newest session is external and not visible to them). The card showed that as "Could not load evaluations from this source. Retry", and under All as "1 place could not be loaded", though retrying can never succeed. A 404 now marks the place hidden: its tab has no count, selecting it says "You cannot see this place's evaluations.", and All neither counts nor flags it. Other failures keep the error and Retry.
• Expired rows say so. A row whose bodies retention stripped has no title or answer, and showed a bare —. It now reads Details expired in the tertiary tone, reusing the drawer's copy.

Testing

• DecisionRecentEvaluations: a 404 place raises no alert, keeps All's rows, has no count, and shows the hidden note; a tab renders the mark markFor returns; an expired row reads Details expired
• DecisionEditorView and the Decision component suites (112); web typecheck and lint
• The 404 and the other two issues were diagnosed on the test console; the mock console has no bot list, so the platform mark itself was checked through the unit test rather than a screenshot

🤖 Generated with Claude Code . Claude Opus 5.5

[agentconnect-md-test]

Approved. The 404 handling matches the conversation evaluation endpoint’s access response, keeps other places’ rows visible, and the new marks and expired-details copy are wired through the card. One minor follow-up: an agent with no custom icon can show a blank mark until the ACP runtime icon registry loads, because AgentIconView may render null after markFor has suppressed the kind glyph fallback. I could not run the focused tests here because this checkout has no installed dependencies.

_{sent by review-bot (Codex · gpt-6-sol) · open in session}

[agentconnect-md-test]

Approved at fc589b6. The added registry check restores the kind glyph while a null-icon agent’s runtime mark is unavailable, and the other reviewed behavior remains unchanged. One minor fallback case remains: an explicit { kind: 'runtime' } icon is truthy here but also renders through AgentMark, so its mark slot can still be blank until the registry loads. I could not run tests in this checkout because dependencies are not installed.

_{sent by review-bot (Codex · gpt-6-sol) · open in session}

[agentconnect-md-test]

Approved at 967646f. The new drawn check covers both null and explicit runtime icons: while the ACP registry has no mark, the Decision card receives no agent mark and keeps its kind glyph; glyph and image icons still render directly. The rest of this PR remains as previously reviewed. Tests could not run in this checkout because dependencies are not installed.

_{sent by review-bot (Codex · gpt-6-sol) · open in session}