Extract JavaScript program metrics from run prompts in audit analysis

[Copilot]

Issue reports from ambient-context runs relied on prompt.txt fallback data but could not quantify embedded JavaScript programs, making it harder to identify repeated high-cost prompt fragments. This change adds language-aware code-block extraction so audit data can report JS-specific prompt overhead directly.

• Prompt analysis model expansion

  • PromptAnalysis now includes:
    • code_blocks_by_language
    • javascript_programs
    • javascript_total_size

• Markdown fenced-code extraction in audit pipeline

  • extractPromptAnalysis now parses fenced blocks from prompt.txt content.
  • Supports both backtick and tilde fences.
  • Normalizes language tags and counts per-language blocks.
  • Aggregates JS payload size for js/javascript blocks only.

• Audit output quality improvement

  • Prompt analysis now surfaces language-level breakdown instead of only total prompt bytes.
  • Enables direct attribution of prompt bloat to repeated JS examples/programs in sampled runs.

analysis := &PromptAnalysis{
    PromptSize: len(data),
    PromptFile: relPromptPath,
}
populatePromptCodeBlockAnalysis(string(data), analysis)
// analysis.JavaScriptPrograms / analysis.JavaScriptTotalSize now populated