peerkit CLI crashes under heavy message workload, and the runner floods logs when retrying

[veeso]

Description

this issue has been automatically created by codex. I still need to verify it.

The peerkit_hole_punch scenario (issue holochain/wind-tunnel#692) was originally meant to send 100 messages of 256 KB to each connected peer. A real smoke test against a local relay showed that this workload crashes one of the peerkit command line tool processes partway through the run. The scenario's runner then keeps retrying the failed agent with no delay between attempts, which turned that one crash into millions of repeated error log lines in under a minute.

As a temporary workaround, the scenario now defaults to sending messages of 32 KB instead of 256 KB (the message count stays at 100 per peer). This keeps the scenario usable, but it no longer tests the full workload the original issue asked for.

Two separate problems need investigating here:

1. Why the peerkit command line tool crashes when asked to send 256 KB messages at this volume, and whether it can be fixed or worked around.
2. Why the wind tunnel scenario runner retries a permanently failed agent with no pause in between, which turns a single crash into a huge burst of repeated error messages.

Dependencies

Related to holochain/wind-tunnel#692.

Notes

Once the crash is understood and fixed (or worked around), the scenario's default message size should be raised back to 256 KB to match the original request in holochain/wind-tunnel#692.

Acceptance criteria

• The cause of the peerkit process crash under the full 256 KB message workload is understood.
• The wind tunnel scenario runner no longer floods the log with repeated errors when an agent fails permanently; it stops or slows down instead.
• The peerkit_hole_punch scenario's default message size is raised back to 256 KB once the above is resolved, and a full run completes cleanly.