fix(report): expose per-case execution metrics - #191

Summary

• distinguish evaluation wall time from per-case tested-agent execution time
• expose per-case input, output, and total tested-agent token usage in JSON, HTML, and Markdown reports
• capture agent-judge duration and token usage separately and add tested-agent, judge, and overall token totals
• keep benchmark metrics compact beside each case heading, including with-Skill, without-Skill, and delta annotations
• preserve the existing total_tokens JSON field for compatibility

Root cause

The top-level report duration comes from EndTime - StartTime, while each case duration comes from the tested-agent session. The HTML report labeled both scopes ambiguously and omitted the per-case token fields that were already present in CaseResult.

Agent-judge sessions also exposed duration and token usage internally, but those metrics were not copied into the report model before JudgeSession was excluded from JSON serialization.

This change is engine-independent. Agent-specific token acquisition, including QoderCLI session parsing, is intentionally out of scope.

User impact

Reports now make Skill execution cost visible without allowing metrics to dominate the response and grading content:

• the header labels total duration as Evaluation wall time
• compact case-heading annotations show with-Skill, without-Skill, and delta metrics
• input/output token details remain available on hover
• agent-judge cost is reported separately
• Markdown and JSON reports use the same metric semantics

Validation

• make fmt
• make verify
• go test -race ./...
• full local benchmark evaluation of expression-calculator using QoderCLI:
  • 2 cases × with/without Skill
  • 3 PASS, 1 baseline FAIL, 0 ERROR
  • evaluation wall time matched Agent + Judge execution time with approximately 0.2s framework overhead
• generated HTML embedded JavaScript syntax and compact-layout assertions

Closes #190