[ASR] Fix GPU memory leak in transcribe_speech.py

[rlangman]

What does this PR do ?

Fix memory leak in utility used by transcribe_speech.py when transcribing partial audio segments, by moving logit tensor stored in results to CPU.

Collection: [ASR]

Changelog

• Modify transcribe_partial_audio to move logits to CPU when when logprobs=False.

Before your PR is "Ready for review"

Pre checks:

• Make sure you read and followed Contributor guidelines
• Did you write any new necessary tests?
• Did you add or update any necessary documentation?
• Does the PR affect components that are optional to install? (Ex: Numba, Pynini, Apex etc)
  • Reviewer: Does the PR have correct import guards for all optional libraries?

PR Type:

• New Feature
• Bugfix
• Documentation

[stevehuang52]

LGTM, thanks~