Since we have the NodeInfo map in known_node_aspects now anyway, use it
for representation changes instead of the current interpreter frame
value swapping mechanism.

We originally didn't want a map for representations, but if we have one
anyway for node types, it makes sense to merge the two mechanisms.

This also allows us to get rid of the "register same as accumulator"
hack for binops, which was trying to avoid redundant representation
changes, since the redundancy is removed by the node aspects mechanism
automatically. It's kept for CheckedInternalizedString, since that isn't
a representation change within our framework (maybe it should be?).

Bug: v8:7700
Change-Id: I92491307f83d3b63fc7a210f6b8754e45d3531a7
Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/3905182Reviewed-by: Jakob Linke <jgruber@chromium.org>
Commit-Queue: Leszek Swirski <leszeks@chromium.org>
Cr-Commit-Position: refs/heads/main@{#83317}