perf: Optimize worker pool and queue concurrency for 32 vCPU

Commit notes

- Worker pool now uses ~75% of available CPUs on high-CPU servers (24 workers on 32 vCPU) - Added environment variable overrides for fine-tuning: - ANALYSIS_POOL_SIZE: Override worker pool size - ANALYSIS_QUEUE_CONCURRENCY: Analysis queue workers (default: 16) - PROCESSING_QUEUE_CONCURRENCY: Audio processing workers (default: 10) - GUEST_QUEUE_CONCURRENCY: Guest sample workers (default: 12) - ASSEMBLY_QUEUE_CONCURRENCY: Chunk assembly workers (default: 8) - Increased max queue size to 5000 for high-throughput scenarios - Reduced block timeout for faster job pickup

On 32 vCPU Railway: - Worker pool: 24 parallel analysis workers - Combined queue concurrency: 46 concurrent jobs - Estimated throughput: 100+ samples/minute

Co-authored-by: armin.naimi <[email redacted]>