Revert "Update R2/KJ to avoid mod-after-put when using ArrayBuffer"

[kentonv]

Reverts #309

This change was merged prematurely, the team hasn't reached consensus on this approach. Some possible concerns:

• What is the performance impact of copying? Is it worth the cost to force copying on existing users?
• Is detaching ArrayBuffers a good API? Will it possibly confuse people who are trying to pass the same ArrayBuffer to multiple calls?

It's possible we'll re-land this exactly the same once we've talked this through.

[jasnell]

I'm good with this or #313.

[kentonv]

I am personally leaning towards #313 as the end state but I'm sufficiently uncertain that I'd really like to poll the team for thoughts first.