Optimize DecodeString for the common case where most of the input isn't
a newline or a padding character.
Also add some testcases found when fuzzing this implementation against
upstream.
Change Decode benchmark to run with different input sizes.

name                 old time/op    new time/op    delta
DecodeString/2-4       71.5ns ± 4%    70.0ns ± 6%     ~     (p=0.246 n=5+5)
DecodeString/4-4        112ns ±25%      91ns ± 2%     ~     (p=0.056 n=5+5)
DecodeString/8-4        136ns ± 5%     126ns ± 5%   -7.33%  (p=0.016 n=5+5)
DecodeString/64-4       872ns ±29%     652ns ±21%  -25.23%  (p=0.032 n=5+5)
DecodeString/8192-4    90.9µs ±21%    61.0µs ±13%  -32.87%  (p=0.008 n=5+5)

name                 old speed      new speed      delta
DecodeString/2-4     56.0MB/s ± 4%  57.2MB/s ± 6%     ~     (p=0.310 n=5+5)
DecodeString/4-4     73.4MB/s ±23%  87.7MB/s ± 2%     ~     (p=0.056 n=5+5)
DecodeString/8-4     87.8MB/s ± 5%  94.8MB/s ± 5%   +7.98%  (p=0.016 n=5+5)
DecodeString/64-4     103MB/s ±24%   136MB/s ±19%  +32.63%  (p=0.032 n=5+5)
DecodeString/8192-4   122MB/s ±19%   180MB/s ±11%  +47.75%  (p=0.008 n=5+5)

Improves #19636

Change-Id: I39667f4fb682a12b3137946d017ad999553c5780
Reviewed-on: https://go-review.googlesource.com/34950Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>