Add pre-v1 robustness, fuzz and soak evidence

[Inczert]

Parent: #209

Context

The existing CI has broad deterministic unit/contract/integration/sanitizer coverage, but the pre-v1 audit found no dedicated fuzz/property or long-running soak campaign.

A stable infrastructure library should have evidence for malformed-input resilience and repeated lifecycle/ownership/reconnect behavior beyond short deterministic fixtures.

Fuzz/property candidates

Prioritize bounded parsers/state machines with externally influenced input:

• VSPW-TP header/frame decode and validation;
• fragmentation/reassembly including overlap/duplicate/conflict cases;
• VSPD protocol codec/record validation;
• CUSE record codec where useful;
• configuration validation/state-machine invariants.

The fuzz targets must remain bounded and suitable for sanitizers. Corpus/regression inputs from discovered failures should be committed where appropriate.

Soak/stress candidates

Add a repeatable long-running campaign for representative virtual paths:

• repeated simulator open/start/traffic/reset/close/reopen;
• sustained full-duplex copied traffic;
• zero-copy acquire/submit/reclaim/release cycling and pool exhaustion/recovery;
• UDP peer restart/reconnect/session rollover under sustained traffic;
• DEVICE/vspwd daemon/client restart cycles on Linux;
• bounded queue pressure without leaks or ownership corruption.

This need not run at maximum duration on every PR. Use a short CI smoke plus a longer scheduled/manual mode if appropriate.

Acceptance

• sanitizer-enabled fuzz/property targets exist for the selected protocol/parser boundaries;
• a committed regression corpus covers any failures discovered;
• short deterministic stress smoke runs in ordinary CI;
• longer soak mode is reproducible and documents duration/seed/configuration;
• repeated runs show no leaks, sanitizer failures, stale ownership, deadlocks or unbounded resource growth;
• results are software robustness evidence, not physical SpaceWire qualification.