Add category-level score summary at end of benchmark run

[ScuttleBot]

Summary

Addresses #41 — adds a summary at the end of benchmark runs that matches the PinchBench website format.

Output Example

================================================================================
🦀 PINCHBENCH SCORE SUMMARY
================================================================================

   Overall Score: 86.0% (19.9 / 23.0)

   CATEGORY               SCORE        TASKS
   --------------------------------------------
   🟢 BASIC              100.0%      1 task
   🟡 CALENDAR            83.0%      1 task
   🟢 CODING             100.0%      1 task
   🟢 COMPREHENSION       93.0%      4 tasks
   🔴 CREATIVE             0.0%      1 task
   ...
   --------------------------------------------
================================================================================

Features

• Overall score — percentage with earned/possible points
• Per-category breakdown — grouped by task category
• Color indicators — 🟢 ≥90%, 🟡 ≥70%, 🔴 <70%
• Task counts — shows how many tasks in each category

Use Case

Users running benchmarks locally with --no-upload can now compare their results directly against the PinchBench leaderboard without digging through verbose output.

Closes #41

[kilo-code-bot]

Code Review Summary

Status: No Issues Found | Recommendation: Merge

Overview

This PR adds a _log_category_summary() function that groups benchmark task scores by category and logs a formatted summary table. The implementation is clean and well-structured:

• Edge cases are handled properly (empty task_entries, missing grading data, empty/missing categories)
• Division-by-zero is guarded with if total_possible > 0 / if data["possible"] > 0 checks
• The task.category attribute defaults to '' in lib_tasks.py, and the code correctly falls back to "UNCATEGORIZED" for falsy values
• The function is purely for logging/display and has no side effects on data flow

Files Reviewed (1 file)

• scripts/benchmark.py - 0 issues