Per-host politeness: rate/parallelism caps

[KingAkeem]

Summary

Add per-host concurrency caps and optional inter-request delay to avoid bans and reduce tail latencies.

Motivation

• Be a good citizen to hosts
• Smooth out heavy-tailed latency distributions

Scope

• Config: --per-host-parallel=N, --per-host-delay=ms
• Implementation via per-host tokens or queues

Acceptance Criteria

• With --per-host-parallel=1, no concurrent requests to the same host
• Delay is respected and measurable in logs/metrics

Tasks

• Implement per-host limiters
• Add tests simulating two hosts with uneven link graphs