Replace ab with locust for realistic load testing

[linconvidal]

Problem

Current stability_test.py uses Apache Bench which sends identical payloads repeatedly:

• Database caching heavily skews results
• Doesn't reveal performance differences based on UTXO count, transaction history, or block size
• Can't vary data per request
• Unrealistic compared to production traffic patterns

Proposed Solution

Tool: Locust (Python-based load testing framework)

Key Benefits:

• Read CSV with multiple rows, cycle through different data per request
• Built-in p95/p99 percentile metrics
• Real-time web UI for monitoring
• Can weight endpoints differently (e.g., account/balance gets 10x more traffic than construction/metadata)
• CI/CD friendly headless mode

Data Strategy

Create varied CSV files with diverse data profiles:

Address Diversity:

• Light (1-10 UTXOs), Medium (100-1K UTXOs), Heavy (10K+ UTXOs)
• New addresses (few txs) vs old addresses (thousands of txs)

Block Diversity:

• Light blocks (1-5 txs) vs heavy blocks (100+ txs)
• Recent blocks vs historical blocks

Transaction Diversity:

• Small (<500 bytes) vs large (>10KB)
• With/without metadata, with/without native assets