feat: Balance review problems with learning new problems

[smithrashell]

Summary

Currently, review problems are picked automatically based on the Leitner system and spaced repetition. However, there's no mechanism to guarantee progress by prioritizing learning new problems alongside reviews.

Current Behavior

• Review problems are selected automatically based on due dates
• New problems are selected to fill remaining session slots
• Heavy review load could potentially crowd out new learning

Desired Behavior

Configurable balance between:

1. Review problems - Problems due for spaced repetition
2. New problems - Problems not yet attempted (learning progress)

Potential Solutions

Option 1: Minimum New Problems Guarantee

• Setting: "Minimum new problems per session: X"
• Ensures at least X new problems in each session
• Reviews fill remaining slots

Option 2: Percentage Split

• Setting: "New vs Review ratio: 60/40"
• 60% of session = new problems
• 40% of session = review problems

Option 3: Review Cap

• Setting: "Maximum review problems per session: X"
• Caps reviews at X, rest are new problems
• Prevents review backlog from blocking progress

Option 4: Smart Balance (Adaptive)

• Algorithm automatically balances based on:
  • Size of review backlog
  • Days since last new problem
  • Current mastery level
  • User's learning velocity

Questions to Consider

• Should this be user-configurable or automatic?
• How does this interact with difficulty progression?
• Should "new" include problems at new difficulty levels vs. truly unseen problems?

Labels

enhancement, algorithm

[smithrashell]

Completed via adaptive approach (Option 4) across multiple commits: 987fd73 (session guard rails + sessionBalancing.js), 385695c (triggered reviews with contextual session composition), and df14935 (removed artificial review ratio sliders). The system now automatically balances review vs new problems based on backlog, mastery level, and contextual relevance.