ExtractSpawnedAgentIDs misses subagents spawned before startLine offset

[khaong]

Problem

ExtractAllModifiedFiles and CalculateTotalTokenUsage both extract spawned agent IDs from the transcript slice starting at startLine. When startLine > 0 (i.e., after a checkpoint), subagents spawned in earlier turns are not discovered.

If those earlier-spawned subagents continue writing to their transcript files in later turns, their file modifications will be missed.

Impact

After a checkpoint is saved (advancing CheckpointTranscriptStart), modifications by previously-spawned subagents become invisible to both file extraction and token usage calculation.

Suggested Fix

Extract agent IDs from the full transcript (startLine=0) while still limiting main-agent modified file extraction to startLine. This ensures all subagent transcripts are scanned regardless of when the agent was spawned.

Context

Identified during review of #323 (ENT-297 fix). This is a pre-existing limitation shared by CalculateTotalTokenUsage.