fix: Use least-tested-first selection for regression testing

[connor-tyndall]

Summary

Fixes #20 - Random testing results in duplicate testing (wasting tokens / time)

• Adds regression_count column to track how many times each feature has been regression tested
• Replaces random selection with least-tested-first ordering
• Ensures even distribution of regression testing across all features

Problem

The current feature_get_for_regression uses func.random() to select features:

.order_by(func.random())

This causes:

• Same features tested repeatedly while others are never tested
• Wasted tokens on duplicate testing
• Incomplete regression coverage

Solution

Track regression test frequency and prioritize least-tested features:

1. New regression_count column - Tracks how many times each feature has been regression tested
2. Least-tested-first ordering - order_by(regression_count.asc(), id.asc())
3. Automatic increment - Counter increases each time a feature is selected

Before

Session 1: Features [5, 12, 8]  (random)
Session 2: Features [5, 3, 12]  (random - duplicates!)
Session 3: Features [5, 8, 5]   (random - more duplicates!)

After

Session 1: Features [1, 2, 3]   (count: 0 → 1)
Session 2: Features [4, 5, 6]   (count: 0 → 1)
Session 3: Features [7, 8, 9]   (count: 0 → 1)
...
Session N: Features [1, 2, 3]   (count: 1 → 2, round-robin)

Test plan

• Verify new databases get regression_count column
• Verify existing databases are migrated correctly
• Call feature_get_for_regression multiple times - verify different features returned each time
• Verify regression_count increments after each call

🤖 Generated with Claude Code

[meirm]

I am using this PR on my server and it works.