Add max_turns_budget experiment to go-logger and support expression-safe Claude max-turns compilation

[Copilot]

This change introduces an A/B experiment for go-logger to compare conservative/standard/generous turn budgets and measure impact on file throughput, compliance, and cost. It wires variant-driven max-turn selection into the workflow while keeping compilation valid and lock output deterministic.

• Workflow experiment wiring

  • Added experiments.max_turns_budget to .github/workflows/go-logger.md with:
    • variants: conservative, standard, generous
    • metric + secondary metrics + guardrails
    • weighted split, start date, analysis type, tags
  • Added variant-driven top-level max-turns expression mapping:
    • conservative -> 20
    • standard -> 35
    • generous -> 55

• Compiler/runtime safety for Claude max-turn expressions

  • Updated pkg/workflow/claude_engine.go to avoid embedding raw GitHub expressions directly in shell run: commands when max-turns is expression-based.
  • For expression values, --max-turns is appended via shell expansion from GH_AW_MAX_TURNS, preserving expression support without template-injection regression.

• Lock regeneration + coverage

  • Recompiled .github/workflows/go-logger.lock.yml to include experiment outputs/spec and mapped max-turn behavior.
  • Added integration coverage in pkg/workflow/max_turns_test.go for Claude top-level expression-based max-turns compilation.

# .github/workflows/go-logger.md
max-turns: "${{ needs.activation.outputs.max_turns_budget == 'conservative' && '20' || needs.activation.outputs.max_turns_budget == 'generous' && '55' || '35' }}"
experiments:
  max_turns_budget:
    variants: [conservative, standard, generous]
    metric: files_successfully_logged_per_run