Add algorithm optimization guide from source code analysis

[Copilot]

Analyzed Twitter's recommendation algorithm codebase to document ranking mechanics, reputation scoring, and visibility optimization strategies.

Changes

New documentation: docs/TWITTER_ALGORITHM_OPTIMIZATION_GUIDE.md

• Ranking mechanics: 4 primary signals (favorites, retweets, replies, dwell time) with configurable weights (-10k to +10k), scored via weighted aggregation
• TweepCred calculation: PageRank-based reputation with age penalties (<30 days), verification boost (score=100), and exponential follower/following ratio penalty when >0.6 with >500 following
• Engagement tracking: In-network vs out-of-network metrics, negative signal weights (reports=-20k, blocks=-1k), USS feature matrix across ML models
• Source references: Extracted from UserMass.scala, Reputation.scala, home-mixer/, scoring weight registries, and signal tracking configs

Updated README.md: Added prominent link to guide

Key algorithm insights

Following/follower ratio penalty from adjustReputationsPostCalculation:

// Exponential penalty for ratio >0.6 with >500 following
divisionFactor = exp(5.0 × (ratio - 0.6))
adjustedScore = originalScore / divisionFactor

// Example: ratio 2.0 = 403× penalty

Score aggregation from RerankerUtil.scala:

finalScore = Σ(engagement_score × weight) + epsilon
// epsilon = 0.001, weights range [-10000, 10000]
// rank_decay = 0.95

Verification gives maximum TweepCred (100) while restricted accounts receive 0.1× multiplier.

---

✨ Let Copilot coding agent set things up for you — coding agent works faster and does higher quality work when set up for your repo.