[LoongArch] Implement a workaround for R_LARCH_PCALA_LO12

It seems R_LARCH_PCALA_LO12 is sometimes used against JIRL even though
such use case is incorrect as per the psABI.

At least, `/usr/lib/loongarch64-linux-gnu/libc_nonshared.a(atexit.oS)`
on `cfarm400.cfarm.net` contains such relocation.

We need this patch to self-host mold on that cfarm machine.

elf/arch-loongarch.cc

@@ -323,7 +323,17 @@ void InputSection<E>::apply_reloc_alloc(Context<E> &ctx, u8 *base) {
       write_k12(loc, (S + A) >> 52);
       break;
     case R_LARCH_PCALA_LO12:
-      write_k12(loc, S + A);
+      // It looks like R_LARCH_PCALA_LO12 is sometimes used for JIRL even
+      // though the instruction takes a 16 bit immediate rather than 12 bits.
+      // It is contrary to the psABI document, but GNU ld has special
+      // code to handle it. We accept it with a warning message.
+      if ((*(ul32 *)loc & 0xfc00'0000) == 0x4c00'0000) {
+        Warn(ctx) << *this << ": invalid use of a R_LARCH_PCALA_LO12 relocation"
+                  << " against a JIRL instruction";
+        write_k16(loc, sign_extend(S + A, 11) >> 2);
+      } else {
+        write_k12(loc, S + A);
+      }
       break;
     case R_LARCH_PCALA_HI20:
       write_j20(loc, hi20(S + A, P));