QVAC-17876 feat[bc]: migrate SDK parakeet transcription to 0.6.0 GGML

[ishanvohra2]

Note: be concise and prefer bullet points.

🎯 What problem does this PR solve?

• SDK parakeet integration still targets pre-0.6 ONNX multi-file loading, which does not match @qvac/transcription-parakeet 0.6.0 single-GGUF GGML addons.
• Apps cannot use parakeet duplex streaming (transcribeStream), Sortformer AOSC knobs, or the new Q8_0 registry models from the SDK.
• Legacy ONNX modelConfig fields fail opaquely instead of guiding migration.

Asana: QVAC-17876

📝 How does it solve it?

• Bump @qvac/transcription-parakeet to ^0.6.0 and load one GGUF via top-level modelSrc (variant auto-detected from GGUF metadata).
• Replace multi-artifact ONNX resolvers with a single model path in the parakeet plugin; reject legacy ONNX modelConfig with LEGACY_PARAKEET_MODEL_DEPRECATED (52210).
• Add parakeet duplex transcribeStream (paced input, partial/final segments, synthetic endOfTurn with source: "parakeet"); wire parakeetStreamingConfig and request lifecycle cancel.
• Discriminate stream endOfTurn by source (whisper vs parakeet); whisper keeps silenceDurationMs, parakeet does not.
• Regenerate model registry with Q8_0 GGUF constants; update examples, docs (0.6.0), and qvac e2e stream suite (desktop smoke validated).

Rebuilds on lessons from reverted #2018 (no TTS bleed-in, structured legacy errors, stream cleanup tests).

🧪 How was it tested?

• cd packages/sdk && bun run lint — pass
• cd packages/sdk && bun run test:unit — pass (parakeet schemas, transcribe VAD/event tests)
• Desktop qvac e2e: parakeet + parakeet-stream suites — 19/19 pass (smoke 4/4)
• Requires @qvac/transcription-parakeet@0.6.0 prebuilds on target platform (darwin-arm64, ios-arm64, etc.)

💥 Breaking Changes

BEFORE:

await loadModel({
  modelType: "parakeet",
  modelConfig: {
    modelType: "tdt",
    parakeetEncoderSrc: PARAKEET_TDT_ENCODER_INT8,
    parakeetDecoderSrc: PARAKEET_TDT_DECODER_INT8,
    parakeetVocabSrc: PARAKEET_TDT_VOCAB,
    parakeetPreprocessorSrc: PARAKEET_TDT_PREPROCESSOR,
  },
});

AFTER:

await loadModel({
  modelType: "parakeet",
  modelSrc: PARAKEET_TDT_0_6B_V3_Q8_0,
  modelConfig: {
    streaming: true,
    streamingChunkMs: 500,
  },
});

Legacy ONNX keys in modelConfig still parse but raise LegacyParakeetModelDeprecatedError with a migration message.

📦 Models

Added models

PARAKEET_TDT_0_6B_V3_Q8_0
PARAKEET_CTC_0_6B_Q8_0
PARAKEET_SORTFORMER_4SPK_V1_Q8_0
PARAKEET_EOU_120M_V1_Q8_0

[Victor-Rodzko]

@ishanvohra2 FYI:

• failed tests on Android not related to current PR - it is a known issue with tests (hopefully it will be fixed here):

All other tests passed on all platforms - GREEN - consider all tests passed!

[github-actions]

QVAC E2E — windows — ❌ failed

Totals: 90/91 passed · 1 failed · 98.9% · 510s
Config: suite=smoke · filter=(none) · exclude=(none)
View run · Artifacts: reports

Results by section

• addon-logging: 1/2 ❌

Failed tests

• addon-logging-during-inference: Test timeout after 120s

[github-actions]

QVAC E2E — ios — ✅ all tests passed (82/91, 1075s)

Config: suite=smoke · filter=(none) · exclude=(none)
View run · Artifacts: reports

[github-actions]

QVAC E2E — linux — ✅ all tests passed (91/91, 238s)

Config: suite=smoke · filter=(none) · exclude=(none)
View run · Artifacts: reports

[github-actions]

QVAC E2E — android — ✅ all tests passed (83/91, 2801s)

Config: suite=smoke · filter=(none) · exclude=(none)
View run · Artifacts: reports · Device Farm logs

[github-actions]

QVAC E2E — macos — ✅ all tests passed (91/91, 313s)

Config: suite=smoke · filter=(none) · exclude=(none)
View run · Artifacts: reports

[ishanvohra2]

/review