hal0 v0.8.2b3

Changes in v0.8.2b3:

• chore(release): bump version to 0.8.2b3 (#978)
• chore(manifest): pin hal0-toolbox-qwen3tts image digest (#977)
• chore: add qwen3tts standalone-to-slot migration runbook + guarded script (#974)
• feat(capabilities): wire voice.tts engine switch (Kokoro CPU / Qwen3-TTS GPU) (#976)
• ci(toolbox): wire hal0-toolbox-qwen3tts build+push to ghcr.io (#975)
• feat(providers): GPU Qwen3-TTS as a hal0-native slot (Qwen3TTSProvider) (#972)
• fix(dispatcher): inject upstream auth headers for remote providers (#973)
• Merge pull request #971 from Hal0ai/docs/benchmarking-ui-handoff
• docs(bench): UI/feature handoff for the benchmarking toolbox
• Merge pull request #969 from Hal0ai/fix/ui-deploy-reliability
• fix(deploy): land UI builds reliably (no-cache index + install to served dist)
• Merge pull request #968 from Hal0ai/ui/gpu-gauge-prefill
• feat(dash): iGPU usage gauge + prefill TTFT readout (yellow ring)
• Merge pull request #967 from Hal0ai/feat/gpu-benchmark-harness
• feat(bench): GPU benchmark harness + hal0-benchctl seam + agent skills
• feat(dash): topbar Kanban/Agent-Chat launchers, global agent chat, Archived lane (#966)

See CHANGELOG.md for full details.