[Orc] Encodes instructions with wrong endianess on aarch64_be

Hi there, it seems that Orc is completely broken on aarch64_be:

Take the following file, `test.ll`:

```ll
define i32 @main(i32 %argc, i8** %argv) {
entry:
ret i32 0
}
```

`lli --force-interpreter test.ll` exits just fine with exit code 0, so does `lli --jit-kind=mcjit test.ll`. But `lli test.ll` results in

```
PLEASE submit a bug report to https://github.com/llvm/llvm-project/issues/ and include the crash backtrace.
Stack dump:
0. Program arguments: lli test.ll
Illegal instruction
```

I believe this is because it emits instructions encoded in big-endian byte ordering rather than little-endian (since aarch64_be uses little endian instruction encoding).

If I run it in gdb with `gdb --args lli test.ll` and run
```
(gdb) x/20i $pc
=> 0xffffedac6054: .inst 0x00021fd6 ; undefined
```

which if we flip the endianess, is `d61f0200`, aka `br x16`.

Reproduce the failure with `lli test.ll`, then compare it with `lli --force-interpreter test.ll` and `lli --jit-kind=mcjit test.ll` on aarch64_be. Start by tracing the Orc JIT path and inspect the emitted instruction bytes in gdb; done means the plain `lli test.ll` run executes successfully with exit code 0 and preserves correct instruction encoding.

Written by the indexing model from the issue text.