fix: prevent fast ascii comparison if char is not letter

[svennergr]

What this PR does / why we need it:
The current containsLower implementation would only work for letters, as the c != s && c+'a'-'A' != s && c != s+'a'-'A' would never work comparing | against\, for example. That will be the case if simple case insensitive regex filters, like |= `(?i)\|` , that should only match against strings that contain |.

This added letter check makes the benchmark worse, but still better than the old version before #15076 was introduced. Maybe there are other ideas of improving?

                                                │   old.txt   │               new.txt                │
                                                │   sec/op    │    sec/op     vs base                │
ContainsLower/short_line_no_match-10              16.27n ± 0%    18.39n ± 1%   +13.06% (p=0.002 n=6)
ContainsLower/short_line_with_match-10            23.91n ± 1%    39.03n ± 2%   +63.22% (p=0.002 n=6)
ContainsLower/long_line_no_match-10               212.2n ± 0%    273.8n ± 4%   +29.03% (p=0.002 n=6)
ContainsLower/long_line_match_start-10            8.937n ± 0%   27.800n ± 1%  +211.07% (p=0.002 n=6)
ContainsLower/long_line_match_middle-10           98.20n ± 0%   124.45n ± 1%   +26.72% (p=0.002 n=6)
ContainsLower/long_line_match_end-10              231.8n ± 1%    305.3n ± 2%   +31.66% (p=0.002 n=6)
ContainsLower/short_unicode_line_no_match-10      47.92n ± 9%    57.74n ± 2%   +20.49% (p=0.002 n=6)
ContainsLower/short_unicode_line_with_match-10    35.47n ± 1%    53.71n ± 1%   +51.42% (p=0.002 n=6)
ContainsLower/long_unicode_line_no_match-10       231.3n ± 3%    274.0n ± 0%   +18.41% (p=0.002 n=6)
ContainsLower/long_unicode_line_match_start-10    233.8n ± 3%    240.4n ± 1%    +2.80% (p=0.037 n=6)
ContainsLower/long_unicode_line_match_middle-10   89.11n ± 1%   111.15n ± 1%   +24.74% (p=0.002 n=6)
ContainsLower/long_unicode_line_match_end-10      335.6n ± 1%    381.9n ± 2%   +13.78% (p=0.002 n=6)
geomean                                           77.39n         105.0n        +35.69%

[cyriltovena]

cc @MasslessParticle contains expects lowered values.

Not sure why we use contains for equality match btw.

[cyriltovena]

LGTM