Fix LinearSolveAutotune printout to use correct benchmark size ranges

[ChrisRackauckas-Claude]

Summary

• Fixed the printout in LinearSolveAutotune.jl to use the actual benchmark size categories instead of hardcoded ranges
• The issue summary now correctly shows results organized by the same sizing used in the benchmark

Problem

The autotune printout was using hardcoded ranges (0-128, 128-256, 256-512, 512+) that didn't align with the actual benchmark size categories defined in get_benchmark_sizes():

• :tiny - 5:5:20
• :small - 20:20:100
• :medium - 100:50:300
• :large - 300:100:1000
• :big - 10000:1000:100000

Solution

Updated categorize_results() in benchmarking.jl to use the proper size ranges with descriptive labels:

• "tiny (5-20)"
• "small (20-100)"
• "medium (100-300)"
• "large (300-1000)"
• "big (10000+)"

Also updated the telemetry formatting in telemetry.jl to:

• Handle the new range format when parsing categories
• Sort ranges in logical order (tiny → small → medium → large → big) instead of alphabetically

Test Plan

Created a test script that verifies the categorization now works correctly with the proper size ranges.

Example output now shows:

#### Recommendations for Float64

| Size Range | Best Algorithm |
|------------|----------------|
| tiny (5-20) | TestAlg |
| small (20-100) | TestAlg |
| medium (100-300) | AlgB |
| large (300-1000) | AlgB |
| big (10000+) | TestAlg |

🤖 Generated with Claude Code