Rely on voice pooling for audio bursts

[supermarsx]

Summary

• remove the cooldown guard from the audio engine so every play request starts a voice while still respecting the pool eviction logic
• deprecate the sound_cooldown_ms configuration option and document the new guidance in the config loader, sample file, and docs
• extend the audio regression test to cover rapid retriggers and adjust config logging coverage for the deprecation warning

Testing

• cmake --preset linux (fails: gtk+-3.0 not found)

---

https://chatgpt.com/codex/tasks/task_e_6906a8948d808325b020c4addae36e5a