[lld] [lld][ELF][x86-64] Do not relax large-model TLS sequences (PR #216194)

[Fangrui Song via llvm-commits]

https://github.com/MaskRay requested changes to this pull request.

Thanks for reporting the lack of support for GCC's large-model TLS GD sequence, but this patch is incorrect.
 
This patch doesn't disable the GD-to-LE optimization for `leaq x at dtpoff(%rax), %rax`, and the `leaq    -0x4(%rax), %rax` is incorrect.

I'll take over and implement this.

https://github.com/llvm/llvm-project/pull/216194