Fix unaligned edge access

[wks]

This allows the OpenJDK binding to load and store ObjectReferences from/to edges at unaligned addresses on x86 and x86_64.

Fixes: #231

[qinsoon]

LGTM. Thanks.

[wks]

https://squirrel.anu.edu.au/plotty/wks/noproject/#0|bobcat-2023-08-21-Mon-135824&benchmark^build^invocation^iteration^mmtk_gc&GC^time^time.stw&|10&iteration^1^4|20&1^invocation|30&1&mmtk_gc&build;build1|41&Histogram%20(with%20CI)^build^mmtk_gc&

I ran lusearch, 100 iterations, 2.5x min heap size, on bobcat.moma. Build1 is master and build2 is this PR.

The difference in STW time is less than 1%. There seems to be a very small amount of performance improvement, but is within the confidence interval.

[wks]

The generated assembly for ProcessEdgesWork::do_work is identical between build1 and build2. I think the difference is a result of noise or subtle differences in unrelated code.

; r14: self; rax: buffer
  109ca1:	49 8b 46 08          	mov    0x8(%r14),%rax      ; let buffer = self.edges.buf;
; rbx: index; r13: edge
  109ca5:	4c 8b 2c d8          	mov    (%rax,%rbx,8),%r13  ; let edge = buffer[index * 8];
  109ca9:	48 ff c3             	inc    %rbx                ; index += 1
; rdx: object
  109cac:	49 8b 55 00          	mov    0x0(%r13),%rdx      ; let object = edge.addr.as_ptr().read_unaligned();
  109cb0:	48 85 d2             	test   %rdx,%rdx           ; if object.is_null() {
  109cb3:	75 cb                	jne    109c80 <mmtk::scheduler::gc_work::<impl mmtk::scheduler::work::GCWork<<E as mmtk::scheduler::gc_work::ProcessEdgesWork>::VM> for E>::do_work+0x30>