Harden autoscaling and add configurable max_workers override

[matburt]

Context: get_auto_max_workers() computes pool size from memory/CPU. We need robust fallbacks and an ops override.

Acceptance criteria

• Guard get_auto_max_workers() against failures from get_mem_in_bytes()/get_cpu_count(); fallback to a safe minimum (≥ 1) with a warning.
• Add optional override setting/env (e.g., DISPATCHERD_MAX_WORKERS); enforce min_workers <= max_workers.
• Unit tests for exception handling, bounds, and override precedence.
• Document knobs in README.md and .env.example.

Touch points

• src/backend/apps/dispatch/pool.py
• src/backend/apps/dispatch/config.py