Release v1.0.0-alpha.10

1.0.0-alpha.9 — CI / supply-chain hardening + voice-frontend maturati…

…on + LLM default flip

Absorbs every change between v1.0.0-alpha.5 and the alpha.9 closeout:
GPU-contention / LLM-warmup fixes, DeepFilterNet capture, half-duplex
post-TTS gate, first-reply latency banner, LLM backend facade,
telegram voice ingestion, voice-optional Cargo feature, full CI /
supply-chain pipeline (fmt + clippy + test, aarch64 cross-compile,
cargo-audit + cargo-deny, shellcheck + ruff), AND the Jetson default
LLM-backend flip to genie-ai-runtime v1.0.0 + Qwen3-4B Q4_K_M
(PR #55, closes #52).

llama.cpp + Phi-4-mini remain the one-line fallback via
[services.llm].backend = "llama_cpp".

Verified end-to-end on Jetson Orin Nano Super 8 GB + LyraT V4.3 +
Phi-4-mini Q4_K_M: ~4 s first-reply latency, accurate sub-second STT
across consecutive cycles.

See CHANGELOG.md for the full breakdown.

1.0.0-alpha.5 — Jetson voice frontend