tray glance: error detail is truncated mid-identifier, and the header count can disagree with the list beneath it

[Dumbris]

Two low-severity findings from live QA of the glance section (#930). Both are cosmetic; neither blocks that PR.

1. Error detail cut mid-identifier

GlanceSection.firstClause splits the error message on the first ., : or newline. Because mcpproxy error text routinely embeds a server:tool identifier, the split lands inside the identifier. Observed:

• Real core message: invalid arguments for memory:create_entities: at '/entities': got string, want array
• Rendered: memory:create_entities · invalid arguments for memory — 8s

It reads as though the server named memory were at fault. This is the common case for mcpproxy errors, not an edge case. The full message is still available in the tooltip.

Suggested fix: skip separators that fall inside a server:tool token, or split on ": " (colon-space) rather than a bare colon.

2. Header count disagrees with the list

Two separate causes, same symptom:

• The header counts all clients while the list caps at 5 rows, with no "+N more" affordance — observed "11 clients" above 5 rows.
• The header count comes from the 30s usage poll while the rows arrive over SSE, so immediately after a call the header can contradict the rows below it — observed header "12 calls this hour" sitting above a 5-second-old row that had already made it 13.

Suggested fix: add a "+N more" row when truncating, and derive the header from the same SSE-updated state the rows use so the two cannot drift.