Render assets from placeholders separately

What we might be able to do is:

In HtmlResponseAttachmentsProcessor::processAttachments() - first of all get the attachments before placeholders are processed, and run the logic from those. Then render placeholders, then get a diff of the libraries, then if there are remaining libraries to render, process those attachments together in a second pass. Then render the header js / css + footer js based on everything we get back. $settings would need to be merged from the two passes into one array.

Put up an initial MR, only did very basic manual testing locally.

Once this works a follow-up can try to run the assetresolver on each individual placeholder and replace #attached there, but one thing at a time.

A few failing tests but views UI and other places tested manually work, so it is mostly test assumptions probably and enough to see what things look like.

However, don't really like the result here. It adds a fair bit of complexity to HtmlResponseAttachmentsProcessor and unless we're able split the aggregates by placeholder (instead of only before/after placeholder rendering) we won't necessarily see much benefit in cache hit rates.

So I think rather than this, we need to look into some way of adding 'hints' to the collection groupers that x libraries should be grouped together. Something like a new asset group 'library_group' that takes an array of libraries, then divide up asset groups by those. Or some kind of virtual/fake library identifying a break between groups of libraries.

Maybe start with real goal and minimize amount of assets? ✨ Support inlining critical CSS for faster core web vitals Active

Then it will be more clear how "to decouple" into placeholders