recovery: fix BBRv2 PROBE_RTT wedge when inflight target is below cwnd floor

[NAVEENKUMARKR777]

Summary

PROBE_RTT can only arm its 200ms exit timer once bytes_in_flight drains
below inflight_target (0.5 × BDP). When that target is below the
connection's cwnd floor (e.g. a large initial congestion window on a
low-RTT path), the floor keeps bytes_in_flight pinned above the target
forever, the exit timer never gets armed, and the connection wedges in
PROBE_RTT indefinitely with cwnd stuck at the floor.

Google's reference implementation (bbr2_probe_rtt.cc) arms the exit
timer when in-flight drops below the inflight target or below the
minimum congestion window, making the check exhaustive over both
cwnd-clamp outcomes:

if (congestion_event.bytes_in_flight <= InflightTarget() ||
    congestion_event.bytes_in_flight <=
        sender_->GetMinimumCongestionWindow()) {
  exit_time_ = congestion_event.event_time + Params().probe_rtt_duration;

The port only kept the first clause. This PR plumbs the connection's cwnd
lower bound (cwnd_limits.min()) through ModeImpl::on_congestion_event
and restores the second clause in ProbeRTT::on_congestion_event.

Fixes #2698.

Test plan

• Added probe_rtt_exits_when_inflight_hits_the_cwnd_floor, a
  regression test that reproduces the wedge (verified it fails
  without the fix and passes with it).
• cargo test -p quiche --features gcongestion,qlog recovery::gcongestion — 117 passed.
• cargo test -p quiche --features gcongestion,qlog,ffi,internal — full suite, 1102 passed + 45 doc-tests.

🤖 Generated with Claude Code