feat(voice): energy-based Voice Activity Detector (#5896)

[hurtdidit]

Summary

• Base branch: master (all contributions)
• What changed and why:
  • Implements EnergyVad, a real energy-based Voice Activity Detector using RMS amplitude analysis, replacing the placeholder NoopVad that always returned Silence.
  • Adds vad_energy_threshold (default 0.01) and vad_silence_timeout_ms (default 500) configuration fields to VoiceDuplexConfig.
  • Wires EnergyVad into VoiceDuplexSession via a new from_config() constructor, so VAD parameters are configurable per-instance.
  • Updates ws.rs to pass config to session creation.
• Scope boundary: Only replaces the VAD implementation — does not change the binary audio frame pipeline, voice event protocol, TTS/STT subsystems, or WebSocket connection handling.
• Blast radius: Minimal. Changes are fully gated behind gateway-voice-duplex feature flag. Existing NoopVad is preserved for backward compatibility. New config fields have safe defaults.
• Linked issue(s): Related [Feature]: Full-duplex voice conversation with barge-in support #5896, Depends on feat(voice): WebSocket binary audio frames with PCM16 validation and VAD pipeline (#5896) #5974 (stacked), Depends on feat(voice): add Vad trait, VoiceEvent protocol, and gateway-voice-duplex feature flag (#5896) #5942 (stacked)

Validation Evidence (required)

All commands run locally on feat/energy-vad branch (commit ed6e4281):

cargo fmt --all -- --check  # passed (no output)
cargo clippy --locked --all-targets -- -D clippy::correctness  # passed (pre-existing warnings in other crates, zero new warnings)
cargo test  # 821+ tests passed

Key test output tails:

# zeroclaw-api (43 tests)
test vad::tests::energy_vad_custom_threshold ... ok
test vad::tests::energy_vad_custom_timeout ... ok
test vad::tests::energy_vad_full_cycle_start_silence_end ... ok
test vad::tests::energy_vad_no_speech_end_before_timeout ... ok
test vad::tests::energy_vad_resets_timeout_on_continuous_speech ... ok
test vad::tests::energy_vad_silence_on_quiet_input ... ok
test vad::tests::energy_vad_speech_end_after_timeout ... ok
test vad::tests::energy_vad_speech_start_again_after_end ... ok
test vad::tests::energy_vad_speech_start_on_loud_input ... ok
test vad::tests::noop_vad_always_silence ... ok
test vad::tests::rms_energy_* ... ok
test result: ok. 43 passed; 0 failed

# zeroclaw-gateway (138 tests)
test voice_duplex::tests::voice_duplex_session_from_config_custom ... ok
test voice_duplex::tests::voice_duplex_session_process_frame_detects_speech ... ok
test result: ok. 138 passed; 0 failed

# zeroclaw-config (553 tests)
test result: ok. 553 passed; 0 failed

# Full suite (lib + bin + component + integration + system)
test result: ok. 230 passed  (lib)
test result: ok. 222 passed  (bin)
test result: ok. 205 passed  (component)
test result: ok. 159 passed  (integration)
test result: ok. 5 passed    (system)

• Beyond CI — what did you manually verify? Manually inspected state machine transitions: Silent→Speaking on loud input, Speaking→Silent after configurable timeout, brief pauses within speech do not prematurely emit SpeechEnd, full cycle (start→continue→end→restart) works correctly. Verified config deserialization with custom thresholds. Verified from_config() creates correct VAD instances.
• If any command was intentionally skipped, why: None skipped.

Security & Privacy Impact (required)

• New permissions, capabilities, or file system access scope? No
• New external network calls? No
• Secrets / tokens / credentials handling changed? No
• PII, real identities, or personal data in diff, tests, fixtures, or docs? No

Compatibility (required)

• Backward compatible? Yes — NoopVad is preserved, new fields have safe defaults via serde defaults, VoiceDuplexSession::new() still works.
• Config / env / CLI surface changed? Yes — Two new optional config fields on [channels.voice_duplex]: vad_energy_threshold (default 0.01) and vad_silence_timeout_ms (default 500). Existing configs without these fields continue to work with defaults.
• Exact upgrade steps for existing users: No action required. New fields are optional with sensible defaults.

Rollback (required for risk: medium and risk: high)

This is a low-risk PR. git revert <sha> is the plan.

• Feature flags or config toggles: Entire change is gated by gateway-voice-duplex feature flag.

i18n Follow-Through

N.A. — No docs or user-facing wording changes.

[Audacity88]

Thanks for the PR. Before I do a full review, could you please update this branch against current master and resolve the merge conflicts?

GitHub currently reports the PR as dirty. The checks on the current head are green, which is helpful, but because this is a risk: high / size: XL voice-duplex change stacked on the related voice work, I would rather review it on a conflict-free head than spend the full pass on a stale merge state.

Once the branch is updated, please refresh the validation evidence in the PR body if anything changes during conflict resolution. The main checks I would expect to still be relevant are:

cargo fmt --all -- --check
cargo clippy --locked --all-targets -- -D clippy::correctness
cargo test

After that I can review the current energy-VAD / voice-duplex integration diff properly.