[lld] [lld][ELF] Sort thunks by their destination to allow quicker convergence (PR #209962)

[Pranav Kant via llvm-commits]

https://github.com/pranavk commented:

@aeubanks I tried it on three large internal binaries which were not converging earlier. They all end up taking less than 10 passes with this patch.

I tried few other sorting techniques.
1. Sort only once during pass 0
2. Sort for every pass
3. Sort at pass == 5

I didn't see much change in number of passes. (3) was marginally better for one large binary.

https://github.com/llvm/llvm-project/pull/209962