Fix silent event loss in DomainParticipant status channel

[alvgaona]

Summary

The DomainParticipantStatusEvent channel silently drops events when its capacity is exceeded. The try_send implementation converts a Full error into Ok(()), so neither the sender nor the consumer can detect the loss. The only indication is a trace!-level log that is invisible at normal log levels.

This PR:

• Increases the channel capacity from 16 to 2048
• Upgrades the overflow log from trace! to warn!
• Minor formatting fixes (comment alignment, import grouping)

How was this discovered

I discovered this bug while building a ROS 2 application that uses RustDDS. When using the status_listener() API to introspect the DDS graph, some nodes' services and topics would not show up at all. Which nodes were visible varied between runs. After ruling out consumer-side issues and confirming SEDP was delivering all data correctly, we traced the problem to the bounded status channel silently dropping events.

Why 16 is not enough

A single DDS participant can expose many endpoints. When SEDP discovers a remote participant, it generates WriterDetected and ReaderDetected events for every endpoint in a burst — faster than the consumer can drain.

For example, a typical ROS 2 node creates ~9 writers and ~7 readers (~16 status events per node). Two nodes produce ~32 events, already exceeding the channel capacity. At 10 nodes (~160 events), 90% of events are silently lost. This is not specific to ROS 2 — any DDS application with multiple endpoints per participant will hit this with modest scale.

Why 2048

• Covers realistic deployments. 2048 handles ~120 participants in a single SEDP burst, which covers the vast majority of DDS systems.
• Negligible cost. The channel is a notification pipe — the actual endpoint data already lives in the DiscoveryDB. At ~200-500 bytes per event, peak buffer is ~1 MB. That's nothing compared to the SPDP/SEDP threads, network buffers, and DDSCache already maintained.
• No performance impact. Send/receive are O(1) regardless of capacity. In steady state the channel is nearly empty — it only fills during initial discovery bursts.
• No behavioral change. Events that previously went through still go through. The only difference is that events that were silently dropped now have room in the channel.

[alvgaona]

@jhelovuo would you mind reviewing this PR? It's a fix for proper interoperability with ROS 2 DDS.

[jhelovuo]

Nice debugging work. Thank you for the contribution.