Update from v3.2 to v4.0 of the Azure AI Vision API

[dkotter]

Description of the Change

In #559, we switched over to using the Azure AI Vision v3.2 API for all Features relying on that. We decided not to switch to the v4.0 of that API as it was still in public preview and had some breaking changes.

That API seems to be more stable now so this PR switches over to that for the following Features:

• Descriptive Text Generator
• Image Tags Generator
• Image Text Extraction

It does not change the following Features:

• Image Cropping: the v4.0 API is fairly different (doesn't actually return a cropped image but returns the image coordinates that need cropped) and will require additional work
• PDF Text Extraction: this doesn't exist in the v4.0 API but has been moved to an entirely new API, Azure AI Document Intelligence, so will look to tackle that in a separate PR

Things to note:

• the v4.0 API supports images up to 20MB, up from previous of 4MB and larger dimensions, up to 16000x16000px
• Image Text Extraction (OCR) used to be two separate API requests. That can all be done in the v4.0 API so the code for this has been simplified (we've removed the OCR class entirely)
• The v4.0 API has less regions supported, in particular for the captions feature, which is used for the Descriptive Text Generator Feature. See https://learn.microsoft.com/en-us/azure/ai-services/computer-vision/overview-image-analysis?tabs=4-0#region-availability for that list
• Also seems the v4.0 API has fixed confidence scores. In v3.0, we recommended a threshold of 70-75%. In v3.2, we saw scores drop to 50-55% and used that as our recommendation. In testing multiple images, it seems 70% is again a good default so that has been updated

In addition, we now output an error message if a valid caption is returned but the confidence score is lower than our threshold. Previously we would just silently discard that, which can lead to people thinking things aren't working. We still don't save that caption but we show an error letting the user know what happened.

Partially closes #827

For some tests, here's some results I got:

Image	v3.2 Caption	v3.2 Confidence Score	v4.0 Caption	v4.0 Confidence Score
A scientist with a microscope	a woman wearing a white coat and white lab coat sitting at a desk	36.85%	a woman in a lab coat and gloves holding a pen and looking at a microscope	70.76%
Stop sign	a stop sign with a cloudy sky	49.51%	a stop sign with clouds in the background	82.06%
A dog	a dog with its mouth open	57.06%	a dog with its mouth open	83.40%

You could argue on if these captions are better or not but they are definitely not worse and the confidence scores are back to being more realistic, which is great as that's an issue that trips up a lot of people

How to test the Change

1. Setup the Descriptive Text Generator Feature with Azure AI. Ensure it works as expected
2. Setup the Image Tags Generator Feature with Azure AI. Ensure it works as expected
3. Setup the Image Text Extraction Feature with Azure AI. Ensure it works as expected

Changelog Entry

Changed - Migrate from the Azure AI Vision v3.2 API to the v4.0 API

Credits

Props @dkotter, @jeffpaul

Checklist:

• I agree to follow this project's Code of Conduct.
• I have updated the documentation accordingly.
• I have added Critical Flows, Test Cases, and/or End-to-End Tests to cover my change.
• All new and existing tests pass.

[iamdharmesh]

Thanks @dkotter. PR looks good to me and it tests well.