chore(deps): update silero requirement from 0.3 to 0.4

[dependabot]

Updates the requirements on silero to permit the latest version.

Changelog

Sourced from silero's changelog.

[0.4.0] - 2026-05-09

Changed

• Breaking — rust-version bumped from 1.85 to 1.88. The new streaming API uses let-chains, which are stable since 1.88.
• Breaking — replaced the closure-based streaming API with a sans-I/O push/pop pattern that mirrors the firered-vad crate.
  • Session::process_stream(stream, samples, |prob| …) -> Result<usize> becomes Session::process_stream(stream, samples) -> Result<&[f32]>. The returned slice borrows from an internal scratch buffer and is valid until the next call.
  • SpeechSegmenter::process_samples(session, stream, samples, |seg| …) becomes SpeechSegmenter::push_samples(session, stream, samples) -> Result<Option<SpeechSegment>>. Pass &[] to drain segments queued by an earlier push (rare, but possible at force-split where one push closes more than one segment).
  • SpeechSegmenter::flush_stream(session, stream, |seg| …) and finish_stream(session, stream, |seg| …) likewise return Result<Option<SpeechSegment>>.
  • SpeechSegmenter::finish_stream no longer resets the segmenter so queued segments survive the call. Call reset() explicitly when you want to reuse the segmenter for a new stream.
  • SpeechSegmenter::pending_segment_count() (new) reports how many segments are still awaiting a push_samples(&[]) drain.
  • SpeechSegmenter::finish no longer resets the segmenter — it enqueues the trailing segment onto pending_segments and pops the head, so undrained segments come out in order before the trailing one. Callers that want to start a fresh stream after finish() must call reset() explicitly.
  • Session::last_probabilities() (new) exposes the slice recorded by the most recent process_stream call. Identical to the slice that call returned on the Ok path; empty after a failed call.
• Atomic streaming inference — Session::process_stream and Session::flush_stream now snapshot *stream at entry and restore it on inference failure. On Err, StreamState is exactly as it was before the call, the pending PCM tail is preserved, and prob_scratch is empty. Callers can retry the same call with the same samples and observe the same result — no risk of StreamState and downstream segmentation drifting apart on transient ORT errors.
• SpeechSegmenter::reset() is no longer const fn — it now clears the new internal pending_segments queue, which uses VecDeque.

Migration

</tr></table> 

... (truncated)

Commits

• See full diff in compare view

Dependabot will resolve any conflicts with this PR as long as you don't alter it yourself. You can also trigger a rebase manually by commenting @dependabot rebase.

---

Dependabot commands and options

You can trigger Dependabot actions by commenting on this PR:

• @dependabot rebase will rebase this PR
• @dependabot recreate will recreate this PR, overwriting any edits that have been made to it
• @dependabot show <dependency name> ignore conditions will show all of the ignore conditions of the specified dependency
• @dependabot ignore this major version will close this PR and stop Dependabot creating any more for this major version (unless you reopen the PR or upgrade to it yourself)
• @dependabot ignore this minor version will close this PR and stop Dependabot creating any more for this minor version (unless you reopen the PR or upgrade to it yourself)
• @dependabot ignore this dependency will close this PR and stop Dependabot creating any more for this dependency (unless you reopen the PR or upgrade to it yourself)

[dependabot]

Looks like silero is up-to-date now, so this is no longer needed.