poly1305: drop broken arm assembly

The ARM assembly uses the reserved G register. This started causing
frequent crashes due to async preemption, but it was already broken in
the presence of signals, including SIGPROF.

name                          old speed      new speed      delta
Chacha20Poly1305/Open-64      2.88MB/s ± 0%  1.85MB/s ± 0%   -35.76%  (p=0.008 n=6+7)
Chacha20Poly1305/Seal-64      3.17MB/s ± 1%  1.97MB/s ± 0%   -37.78%  (p=0.000 n=10+8)
Chacha20Poly1305/Open-64-X    2.41MB/s ± 0%  1.61MB/s ± 0%   -33.29%  (p=0.000 n=9+9)
Chacha20Poly1305/Seal-64-X    2.55MB/s ± 0%  1.64MB/s ± 0%   -35.61%  (p=0.000 n=10+9)
Chacha20Poly1305/Open-1350    8.43MB/s ± 0%  4.15MB/s ± 0%   -50.78%  (p=0.000 n=10+10)
Chacha20Poly1305/Seal-1350    8.55MB/s ± 0%  4.18MB/s ± 0%   -51.12%  (p=0.000 n=9+9)
Chacha20Poly1305/Open-1350-X  8.16MB/s ± 0%  4.06MB/s ± 0%   -50.18%  (p=0.000 n=10+10)
Chacha20Poly1305/Seal-1350-X  8.24MB/s ± 1%  4.08MB/s ± 1%   -50.53%  (p=0.000 n=10+10)
Chacha20Poly1305/Open-8192    9.73MB/s ± 1%  4.56MB/s ± 0%   -53.15%  (p=0.000 n=9+10)
Chacha20Poly1305/Seal-8192    9.57MB/s ± 0%  4.52MB/s ± 0%   -52.77%  (p=0.000 n=9+9)
Chacha20Poly1305/Open-8192-X  9.65MB/s ± 0%  4.54MB/s ± 0%   -52.95%  (p=0.000 n=10+7)
Chacha20Poly1305/Seal-8192-X  9.47MB/s ± 1%  4.50MB/s ± 0%   -52.50%  (p=0.000 n=10+9)

Fixes golang/go#35511

Change-Id: I5e5ca3a0499f04c5fece5bc669a417e32d2656c6
Reviewed-on: https://go-review.googlesource.com/c/crypto/+/213880
Run-TryBot: Filippo Valsorda <[email protected]>
TryBot-Result: Gobot Gobot <[email protected]>
Reviewed-by: Brad Fitzpatrick <[email protected]>
Reviewed-by: Cherry Zhang <[email protected]>

Author: FiloSottile