fix: correct minimap bucket category resolution

[lukecotter]

📝 PR Overview

Fixes incorrect category resolution in minimap density buckets by reverting a binary search optimization that caused a performance regression.

The minimap was using binary search to find frames per bucket on-demand, but this approach was slower (~90ms) than the original single-pass implementation (~10-20ms). This PR reverts to collecting frames during iteration for correct and fast category resolution.

🛠️ Changes made

• Reverted binary search frame lookup to single-pass frame collection
• Restored framesPerBucket array for skyline algorithm
• Removed getFramesInRange() binary search helper
• Updated performance documentation to reflect actual measurements

🧩 Type of change (check all applicable)

• 🐛 Bug fix - something not working as expected
• ⚡ Performance Improvement

🔗 Related Issues

N/A - Performance regression from recent optimization

✅ Tests added?

• 👍 yes - added comprehensive minimap density tests

📚 Docs updated?

• 🙅 not needed - inline documentation already updated

Anything else we need to know?

Performance measurements:

• Before: ~90ms (binary search overhead from repeated searches)
• After: ~10-20ms (single O(N) pass as originally intended)

The binary search optimization attempted to avoid storing frame references, but the repeated search overhead outweighed any memory benefits.