Pin Qualcomm HF models to a working version

[petewarden]

Description

It seems like Qualcomm changed the file and input/output names for their HuggingFace Whisper models in early June 2025. This causes the Android example app to crash, since it's pulling from the top of tree, but expects the models as they were before QC's changes. This PR adds an option to pin models to a particular revision, and sets those revisions to be the commits that were current in May 2025 for QC's tiny and base models. This solves the crash.

Type of Change

• Bug fix 🐛
• New feature 🚀
• Refactor 🔄
• Documentation update 📖
• Other (please describe)

Test Plan

• I have run bash test/test_build_all.sh and it ran successfully
• I have tested this change on all relevant platforms.

Checklist

• My code follows the project's style guidelines.
• I have updated relevant documentation (if applicable).
• I have added appropriate tests (if applicable).
• I have self-reviewed my code before requesting review.

[chen-argmax]

Hi @petewarden thanks very much for providing the fix! 👍
We are trying to have similar API interfaces for downloading from huggingface across Android and iOS, specifically exposing revision as a separate parameter to the downloading functions instead of adding it to Repo(reference from swift).

I have created this PR on top of yours, could you take a look and update this PR?

[petewarden]

Thanks @chen-argmax ! I've integrated your changes.

[chen-argmax]

verified locally, lgtm! 👍

[chen-argmax]

@petewarden
oops the detekt failed, created another PR to fix it