cmd/compile/internal/ssa: a wrong arm64 rule

[benshi001]

In the file cmd/compile/internal/ssa/gen/ARM64.rules, line 1468 - 1498.

The rule seems to combine 8 load-byte to a single LE load dword. However the final REV is wrong, since it is a LE load.

I will correct it along with my CL touching that part.

[gopherbot]

Change https://golang.org/cl/107376 mentions this issue: cmd/compile: optimize ARM64 with register indexed load/store