The tests were assuming that the destructor leaves behind memory in a
defined state when the object was allocated with placement new. Turns
out gcc with no component builds optimizes away the resetting of the
memory.

There's a simpler way to test the functionality by inspecting global
handle counts.

Bug: v8:9639, chromium:995684
Change-Id: I253d84910414c62ca314507b20d2c819f925ea6e
Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/1762512
Commit-Queue: Michael Lippautz <mlippautz@chromium.org>
Reviewed-by: Ulan Degenbaev <ulan@chromium.org>
Cr-Commit-Position: refs/heads/master@{#63303}