Figure out why images seem to be getting sent for translation despite logic to strip them out

[arsalansufi]

Not a huge deal but could be relevant for our VxMark pilot town (New London) if they have an image in their ballot measure. Worth getting ahead of accordingly

https://votingworks.slack.com/archives/C085YT4798C/p1738349866764009

[eventualbuddha]

Maybe in whole or in part because we don't actually use <img> tags or src= for images, since we convert them to SVG here.

vxsuite/apps/design/frontend/src/image_input.tsx

Lines 92 to 97 in 9f44f93

	async function bitmapImageToSvg(imageDataUrl: string) {
	const { width, height } = await getBitmapImageDimensions(imageDataUrl);
	return `<svg xmlns="http://www.w3.org/2000/svg" width="${width}" height="${height}" viewBox="0 0 ${width} ${height}">
	<image href="${imageDataUrl}" width="${width}" height="${height}" />
	</svg>`;
	}

So, we probably should actually update both locations to look for SVG instead:

vxsuite/libs/backend/src/language_and_audio/translator.ts

Lines 66 to 75 in 9f44f93

	// Google Cloud will preserve HTML tags fairly well, so we can pass HTML
	// rich text directly to the API. However, it has a max string length limit,
	// so base64 encoded img src attributes are generally too long to include.
	// We strip them out in order and replace them after translating.
	const srcRegex = /src="([^"]*)"/g;
	const srcAttrsArray = textArray.map((text) =>
	iter(text.matchAll(srcRegex))
	.map((match) => match[0])
	.toArray()
	);

vxsuite/libs/backend/src/language_and_audio/speech_synthesizer.ts

Line 135 in 9f44f93

return { textContent: '[image]' };

[arsalansufi]

Confirmed that New London doesn't have an image in their ballot, so moving to Medium pri. Still want us to get to it, but it's not blocking for this ballot production + machine configuration run.