feat: add OCR for image descriptions using tesseract.js

[nolanlawson]

Another feature shamelessly stolen from Mastodon.

I'm not super excited or satisfied with this so far, but it's a good accessibility feature so I think it's worth having.

Problems:

• tesseract doesn't play nice with CSP, this required hacks to work around Add option not to use a blob: URL for the worker source naptha/tesseract.js#322
• tesseract has a bug with broken/failed image URLs If an image URL fails to fetch, the TesseractWorker silently fails forever naptha/tesseract.js#325
• tesseract doesn't seem to support WASM yet; I have to use the wasm.js file which is ~1MB bigger
• some instance admins may not have enabled CORS for media assets (I hadn't). If so this feature won't work
• it's a big dependency and really awkward to manage like 3 separate files using file-loader and other weird hacks

Overally though I'm willing to move forward with this; hopefully tesseract will get better when it's out of alpha/beta.

[nolanlawson]