CustomVoice TTS v0.7.1 W8A16 — radxa SenseVoice ASR semantic-match, ~…

…49% memory savings

Validated on Orin NX 2026-05-26:
- engine: 435 MB (vs 862 MB FP16, -49.5%)
- Peak Unified Memory: 1026 MB (vs 2001 MB FP16, -48.7%)
- ASR: '今天天气真不错哦。' (FP16: '今天天气真不错。', semantic identical)
- W8A16 WAV md5: 7c4e1825ca1bcbaa4a9b61505967e79d
- Same patched binary as FP16 tag — engine quantization decoupled

/goal 3/3 sub-goals achieved.
Memory: customvoice_tts_w8a16_PASS_2026_05_26.md

CustomVoice TTS v0.7.1 fork-port FP16 baseline — radxa SenseVoice ASR…

… byte-exact '今天天气真不错。'

Validated on Orin NX 2026-05-26:
- binary md5: f50fedc960d8edf7304f897cddbbdaf7
- plugin md5: 3d6761ebbe0946720f9c1d35a56c1cda
- golden wav md5: a4f56dcbccc580a58a9bab0e0b727eae (2.08s 24kHz mono)
- orin-nx snapshot: ~/customvoice-v071-snapshot/20260526/

3-bug root cause: see docs/specs/customvoice-tts-fork-port-handoff.md
Memory: customvoice_tts_fork_port_PASS_2026_05_26.md

Merge pull request NVIDIA#90 from nvxingkaiz/dev-release/0.7.1

TensorRT Edge-LLM 0.7.1 Release

Merge pull request NVIDIA#76 from nvxingkaiz/dev-release/0.7.0

TensorRT Edge-LLM 0.7.0 Release

Merge pull request NVIDIA#69 from nvluxiaoz/dev-release/0.6.1

TensorRT Edge-LLM 0.6.1 Release

Merge pull request NVIDIA#53 from nvluxiaoz/dev-release/0.6.0

TensorRT Edge-LLM 0.6.0 Release Patch

Merge pull request NVIDIA#34 from nvluxiaoz/dev-release/0.5.0

TensorRT Edge-LLM 0.5.0 Release

Merge pull request NVIDIA#3 from nvluxiaoz/dev-release/0.4.0

TensorRT Edge-LLM 0.4.0 release