runtime: using fallback hash unconditionally

[randall77]

First reported in #37086 by @SlinSo:

i am not sure if this is related, but i saw some differences in map access performance, too. In my case I am not using interface{} as a key, but instead a map[int]int

https://github.com/SlinSo/random/blob/master/mapaccess_test.go

pprof excerpt:
go1.13.8:
0.73s 9.53% 82.64% 0.73s 9.53% runtime.aeshash64

go1.14rc1:
1230ms 15.81% 66.20% 1410ms 18.12% runtime.memhash64Fallback
140ms 1.80% 97.43% 140ms 1.80% runtime.memhash64

We should not be using the fallback hash unless the processor can't do the AES instructions (which, as evidenced by the 1.13 results, it has).

[gopherbot]

Change https://golang.org/cl/219337 mentions this issue: runtime: fix fallback logic for aeshash on 32/64 bit