Before this change ivy didn't properly bootstrap components that used selectors more complex than a tag name. The crux of the issue is that we were trying to use the selectors property of a component definition in 2 different code paths:

• bootstrap;
• dynamic instantiation.

It tuns out that selector should be interpreted differently for those 2 scenarios:

• bootstrap: we need a selector as string (as written by a directive author);
• dynamic instantiation: we need a tag name only (defaulting to div if a given selector doesn't contain tag name).

This PR tries to fix the identified issue by clearly separating 2 usage scenario (and hence having 2 separate fields on a ComponentFactory: selector and tagName). Sadly this PR has one obvious issue - it doesn't fully recreate a selector's string (as written by a user) - to do it properly we would have to add more code (non tree-shakable) to stringify a parsed selector in all cases.

So it seems like we need to decide on the following trade-offs:

• correctness (as in this PR);
• fixed code size in the framework (fully re-create a selector string from its parsed version);
• generated code size (we could add oryginal selector string to a directive definition).

I'm leaning on the "be correct and add more code to the framework) side, but opening a draft PR so we can discuss.

Fixes #34349