[lld] [LLD][ELF] Do not reuse thunks in OVERLAYs (PR #200415)

[Peter Smith via llvm-commits]

smithp35 wrote:

@Gelbpunkt  I've made https://github.com/llvm/llvm-project/pull/222565 which should reduce the memory usage of the test without affecting the expected outcome. Can you give that a try?

I didn't get any OOM notifications from the 32-bit Arm buildbots, although these will be 32-bit containers running on a 64-bit kernel so will get 4 GiB of memory, and likely have access to swap too.

https://github.com/llvm/llvm-project/pull/200415