THe change in https://chromium-review.googlesource.com/695108 flushed
out an issue with the IC::UpdatePolymorphicIC logic, where we'd try to
stay MONOMORPHIC or POLYMORPHIC as long as the internalized name doesn't
change. But the calling code already does the internalization for keyed
accesses with Strings, so we need to double check that the same
combination of (map, handler) is not already in the list, and properly
go to MEGAMORPHIC state if there's such a pair already.

This seriously tanked the six-speed-object-literals-ext-es5.js benchmark
on AWFY.

Bug: v8:6367, v8:6278, v8:6344
Change-Id: I90ea88d1fe61c165990c0a10d4a8687ffe351986
Reviewed-on: https://chromium-review.googlesource.com/695307Reviewed-by: Jakob Kummerow <jkummerow@chromium.org>
Commit-Queue: Benedikt Meurer <bmeurer@chromium.org>
Cr-Commit-Position: refs/heads/master@{#48268}