Pacing based congestion controller (BBR-lite) alongside the sliding window

[Segfaultd]

Summary

Add a modern pacing based congestion controller as a third implementation behind the existing congestion control abstraction.

Motivation

CCRakNetSlidingWindow is a loss based AIMD design from the 2000s. On modern bufferbloated links it fills queues and inflates latency, and on high bandwidth-delay-product links it ramps too slowly. Delay aware, pacing based designs (BBR style) hold latency down while using available bandwidth, which matters more for games than raw throughput.

Proposal

• Implement a BBR-lite controller: continuous bandwidth and min RTT estimation, pacing gain cycling, and a send rate target instead of a pure window
• Slot it in alongside CCRakNetSlidingWindow and CCRakNetUDT behind the existing compile time selection in NativeFeatureIncludes.h, ideally promoted to a runtime choice per peer
• Emit pacing decisions as a target batch size per tick so the send path can size its datagram batches from the same signal

Validation

• Unit tests for the estimator math (bandwidth sampling, min RTT windows, pacing gain schedule) as pure logic tests
• Integration comparison under tc netem profiles (loss, delay, bufferbloat) in a Linux container: new controller should match or beat the sliding window on throughput while keeping queueing delay materially lower
• No regression in the existing reliability integration suites with the new controller enabled

Acceptance criteria

• Controller selectable without recompiling dependent code
• Documented guidance on when to choose which controller