fix(tracing): clarify interrupted-span message so restarts aren't read as agent failures

Pre-release review (End User + Chaos Gremlin personas) flagged that the
`statusMessage` set on in-flight generation spans during `rehydrateFromFile`
reused internal jargon ("cache eviction before step-finish") and read like a
tool/agent bug. On a postmortem surface, a span reconstructed after a worker
restart must not be mistaken for a real agent failure.

Reword to plain language that names the cause and explicitly disambiguates it
from an agent failure. The span keeps `status: "error"` (the boundary stays
visible in the waterfall) and the message still contains "interrupted", so the
rehydrate behavioral test contract is unchanged.

Edit is inside the existing `altimate_change` block (tracing.ts:499-600).

Author: anandgupta42

packages/opencode/src/altimate/observability/tracing.ts

@@ -591,7 +591,7 @@ export class Trace {
       if (s.kind === "generation" && s.endTime === undefined) {
         s.endTime = now
         s.status = "error"
-        s.statusMessage = "interrupted (worker restart / cache eviction before step-finish)"
+        s.statusMessage = "interrupted — altimate-code restarted before this step finished recording; not an agent failure"
       }
     }
     this.endTraceStarted = false