feat(transcription): add drag-and-drop file transcription tab

[sperka]

What

A new Transcribe tab: import audio/video files by drag-and-drop or file picker and transcribe them with the currently selected model. Adapted from kitlangton#232 for this fork's dual-engine (batch + streaming) setup.

Works with every engine

The reference PR sent all files through transcription.transcribe(url:...), but this fork's Nemotron streaming model rejects file-based transcribe (it decodes live). So:

• WhisperKit / Parakeet batch - use the existing file-based transcribe.
• Nemotron streaming - new NemotronStreamingClient.transcribeFile(...) one-shot path: feeds the whole file through the streaming manager (process(audioBuffer:) + finish()) in ~5s blocks. TranscriptionClient.transcribe(url:) now routes streaming models here instead of throwing.

The feature calls one unified transcribe(url, model, ...) for all three engines - no engine special-casing in the UI.

Per-job progress + elapsed time

• Determinate % + bar for the Nemotron file path (real framesFed/total from the block-feed loop).
• Indeterminate bar for Whisper/Parakeet, which don't expose a reliable decode fraction (no invented percentage that would stall or jump).
• Live elapsed timer on every job (M:SS), frozen at the final value on completion.

Also

• Multi-file import, per-job copy-to-clipboard / remove, "Clear Finished".
• Clear errors for unsupported or silent-video files.
• Imported transcripts save to History (preserving the original file extension) when history is enabled.

Testing

xcodebuild -scheme Hex -configuration Debug -skipMacroValidation CODE_SIGNING_ALLOWED=NO -quiet build
cd HexCore && swift test

Both pass (build clean; 62 tests / 5 suites). Runtime-verified across all three engines with live progress + elapsed timer.

Changeset

.changeset/81d5e24d.md (minor).