fix: keep route depth within queue capacity

[Raphjacksun7]

Purpose

Prevent route metrics from reporting a queue depth above the route's configured capacity when producer and consumer counters are sampled concurrently.

Product workflow

The installed Python slow-consumer gate fills an eight-frame route while capture continues. Windows exposed an impossible peak value even though the native ring buffer rejected frames at its configured capacity. This patch makes the observation match the state the queue can actually hold.

Architecture

The native ring buffer remains the queue implementation. Capture, routing, delivery policy, and callback behavior are unchanged. Only the reported current and peak depth are capped by the configured route capacity.

Verification

• focused delayed-counter regression
• 500 Core unit tests
• all integration, ABI, example, and benchmark targets
• strict Clippy
• Rustdoc with warnings denied
• CODE_PROTOCOL check
• package and publish dry run for 1.1.7

Remaining proof limits

This fixes an observation invariant. It does not add Windows physical-device latency evidence or change the 10 ms and 20 ms capture profiles.

Scaffold inventory

No scaffold, mock, or loopback-only implementation was added.

Decision

PASS pending the required GitHub checks.