The LoongArch64 port is looking good based on the integration tests (and the capability to bootstrap Wild itself!). I would like to list a couple of significant changes/needs for the port:

1. The port needs a couple of High/Low part relocations where the former needs a bias (typically 0x800), so I decided to make it a first class citizen in the RelocationKindInfo type (plus I noticed a couple of relocations need to first apply addend before masking -> that's why I touched so many relocation expression calculations).
2. LoongArch64 does not use a specific TLSGD Low part relocations for GOT and so we need to special-case it a bit.
3. The target also uses the page masking, so I decided to make it an explicit parameter, rather than the currently used AArch64 constants.
4. There's one super-complex relocation calculation that I decided to move to a separate function highest_relocation_with_bias.
5. Just recently, there was a new addition of relocation in the latest spec release (v2.50) - for now I haven't implemented them.
6. PairSubtractionULEB128 has been factored out as it's something used by both AArch64 and the newly added LoongArch64 port.

Some early benchmarks for linking the Mold project on Loongson-3A6000 8 core CPU:

| Linker | With debug info (s) | Without debug info (s) |
|--------|----------------------|-------------------------|
| bfd    | 65.0                 | 24.0                    |
| mold   | 4.4                  | 1.82                    |
| lld    | 10.4                 | 8.0                     |
| wild   | 11.8                 | 0.953                   |

I would like to thank @Gelbpunkt for the SSH access to the Alpine LoongArch64 machine! Really helped me.

Resolves: #1381