Make remote Eve startup and fallback resilient

[ProfEngel]

Summary

• let the Windows remote Eve client start independently, retry unavailable Ubuntu hosts, discard stale audio, and hand the microphone back to Legacy during outages
• reserve the single realtime pipeline only for the Bridge-selected speaker so Desktop and Companion devices can hand Eve over without loading duplicate GPU pipelines
• clear stale handoff markers and terminate the complete Windows venv process tree during Voice fallback
• let the Ubuntu GPU server become ready without waiting for the remote Windows Core warm-up, preserve deployment-specific GGUF quantization, and allow longer cold starts
• document independent Windows/Ubuntu startup and private-network deployment

Safety

• no tokens, credentials, private endpoints, host usernames, or machine-specific configuration are included
• runtime payload/state files and the local production runbook remain outside the branch
• existing Voice quantization remains opt-in through TRINITY_QWENTTS_QUANT

Validation

• python -m compileall -q core agents RAG trinity_launcher.py
• python -m pytest -q — 324 passed
• live restart/fallback checks verified Core and Bridge availability, remote reconnect, Legacy microphone recovery, and single-device pipeline handoff