Speed up length of text strings

[01mf02]

This is an improved version of #394 which does not use core::str::from_utf8, but rather implements the same technique as in BurntSushi/bstr#223. That should allow it to achieve much more stable results for strings that contain a few invalid UTF-8 sequences, but which are still mostly valid UTF-8.

Furthermore, this removes a few casts for length calculation, which previously could truncate the length of very large strings.

Results:

$ hyperfine -L v main,length-str2 "target/release/jaq-{v} -n '[\"\" | limit(100000; recurse(. + \"a\")) | length] | add'"
Benchmark 1: target/release/jaq-main -n '["" | limit(100000; recurse(. + "a")) | length] | add'
  Time (mean ± σ):      2.495 s ±  0.009 s    [User: 2.481 s, System: 0.003 s]
  Range (min … max):    2.487 s …  2.512 s    10 runs
 
Benchmark 2: target/release/jaq-length-str2 -n '["" | limit(100000; recurse(. + "a")) | length] | add'
  Time (mean ± σ):     156.4 ms ±   1.6 ms    [User: 152.4 ms, System: 3.1 ms]
  Range (min … max):   154.6 ms … 162.1 ms    19 runs
 
Summary
  target/release/jaq-length-str2 -n '["" | limit(100000; recurse(. + "a")) | length] | add' ran
   15.96 ± 0.17 times faster than target/release/jaq-main -n '["" | limit(100000; recurse(. + "a")) | length] | add'

In the pure ASCII case, calculating string length is now up to 16 times faster than in main.