Medium string hashing inline(never) instead of cold

[hoxxep]

Performance improvement for strurl -4%. Spotted while optimising foldhash. Needs a full benchmarking run.

TODO:

• Non-portable read_u64 and read_u32

[hoxxep]

Results suggest 3% improvement to struuid, 5% improvement to strdate, 1.5% improvement to strenglishword and strurl.

For quality hasher, it's a much bigger improvement, as the smaller write code size seems to allow inlining the Hash::hash(str) call more often, and can improve performance by around 10-20% on various string lengths.

realworld/StrUuid/hashonly-struuid-rapidhash-q                                                                             
                        time:   [3.2958 ns 3.2980 ns 3.3003 ns]
                        change: [-12.157% -11.598% -11.182%] (p = 0.00 < 0.05)
                        Performance has improved.
realworld/StrUuid/hashonly-struuid-rapidhash-f                                                                             
                        time:   [2.4988 ns 2.5025 ns 2.5075 ns]
                        change: [-3.4689% -3.0047% -2.6492%] (p = 0.00 < 0.05)
                        Performance has improved.

realworld/StrDate/hashonly-strdate-rapidhash-q                                                                             
                        time:   [2.0761 ns 2.0795 ns 2.0832 ns]
                        change: [-12.625% -12.231% -11.893%] (p = 0.00 < 0.05)
                        Performance has improved.
realworld/StrDate/hashonly-strdate-rapidhash-f                                                                             
                        time:   [1.3631 ns 1.3648 ns 1.3668 ns]
                        change: [-5.8023% -5.6102% -5.4214%] (p = 0.00 < 0.05)
                        Performance has improved.

realworld/StrEnglishWord/hashonly-strenglishword-rapidhash-q                                                                             
                        time:   [2.1668 ns 2.1692 ns 2.1718 ns]
                        change: [-23.959% -23.245% -22.518%] (p = 0.00 < 0.05)
                        Performance has improved.
realworld/StrEnglishWord/hashonly-strenglishword-rapidhash-f                                                                             
                        time:   [1.4604 ns 1.4627 ns 1.4654 ns]
                        change: [-1.6729% -1.4254% -1.1751%] (p = 0.00 < 0.05)
                        Performance has improved.

realworld/StrUrl/hashonly-strurl-rapidhash-q                                                                             
                        time:   [5.6554 ns 5.7057 ns 5.7562 ns]
                        change: [-12.927% -11.788% -10.725%] (p = 0.00 < 0.05)
                        Performance has improved.
realworld/StrUrl/hashonly-strurl-rapidhash-f                                                                             
                        time:   [4.7906 ns 4.8248 ns 4.8587 ns]
                        change: [-2.4629% -1.5265% -0.4834%] (p = 0.00 < 0.05)
                        Change within noise threshold.

realworld/Kilobyte/hashonly-kilobyte-rapidhash-q                                                                             
                        time:   [27.642 ns 27.670 ns 27.697 ns]
                        change: [-9.9109% -9.7570% -9.6045%] (p = 0.00 < 0.05)
                        Performance has improved.
realworld/Kilobyte/hashonly-kilobyte-rapidhash-f                                                                             
                        time:   [26.783 ns 26.829 ns 26.888 ns]
                        change: [-1.3766% -1.1587% -0.9183%] (p = 0.00 < 0.05)
                        Change within noise threshold.

realworld/TenKilobyte/hashonly-tenkilobyte-rapidhash-q                                                                            
                        time:   [233.21 ns 233.75 ns 234.35 ns]
                        change: [-0.6903% -0.3787% -0.0557%] (p = 0.02 < 0.05)
                        Change within noise threshold.
realworld/TenKilobyte/hashonly-tenkilobyte-rapidhash-f                                                                            
                        time:   [232.82 ns 233.23 ns 233.65 ns]
                        change: [-0.0684% +0.2162% +0.4645%] (p = 0.10 > 0.05)
                        No change in performance detected.