refactor(ivy): merge directives into LViewData - #26316

This PR flattens the LViewData[DIRECTIVES] array into LViewData and the TView.directives array into TView.data. This reduces memory pressure by avoiding the creation of unnecessary nested arrays. It also implements the host binding algorithm outlined in VIEW_DATA.md.

Notes:

• loadDirective() was being used in hostBinding and contentQueries functions as a special instruction to read from the directives array. Now that there isn't a directives array, we can simply use the existing load() instruction in these cases, so the loadDirective() instruction has been removed.
• We were also using loadDirective() in a number of places as a testing utility. This can't really be replaced with load() because now that directives are flat, it's harder to predict where the directive will live in viewData (e.g. there might be other directives with injectors and host bindings ahead of it), so I've created a new testing utility to replace it called getDirectiveOnNode(). This takes the node index and an optional directive offset.
• We were previously creating a node injector for any node that injected special tokens, like ElementRef. This is unnecessary given that they are created using factories, so injecting these tokens no longer triggers the creation of an injector on its own.
• We were also creating a node injector for directives private to DI if they happened to inject something else. This also isn't correct since the injector it would create would be empty. Now nodes with only private directives that have injectables will simply skip to the parent injector.

For more information about this change, see VIEW_DATA.md.