using a table-based convert_hex_to_binary function for better speed

[lemire]

This is an alternative to #473

Using the search-params benchmark, we can use benchmarks/bench_search_params to test the speed at which we can parse search parameters. This task is percent-decoding intensive.

GCC 11, Ice Lake processor.

Main:

----------------------------------------------------------------------------------- Benchmark Time CPU Iterations UserCounters... ----------------------------------------------------------------------------------- url_search_params_AdaURL 65900 ns 65735 ns 10671 GHz=3.19635 cycle/byte=11.6984 cycles/url=32.773k instructions/byte=40.3667 instructions/cycle=3.45062 instructions/ns=11.0294 instructions/url=113.087k ns/url=10.2532k speed=170.473M/s time/byte=5.86604ns time/url=16.4337us url/s=60.8505k/s 

PR #473

----------------------------------------------------------------------------------- Benchmark Time CPU Iterations UserCounters... ----------------------------------------------------------------------------------- url_search_params_AdaURL 65202 ns 65028 ns 10785 GHz=3.19652 cycle/byte=11.5858 cycles/url=32.4575k instructions/byte=40.3469 instructions/cycle=3.48245 instructions/ns=11.1317 instructions/url=113.032k ns/url=10.154k speed=172.325M/s time/byte=5.80298ns time/url=16.257us url/s=61.5118k/s 

Table approach:

----------------------------------------------------------------------------------- Benchmark Time CPU Iterations UserCounters... ----------------------------------------------------------------------------------- url_search_params_AdaURL 57930 ns 57759 ns 11989 GHz=1.99858 cycle/byte=9.17089 cycles/url=25.6923k instructions/byte=40.4051 instructions/cycle=4.40579 instructions/ns=8.80533 instructions/url=113.195k ns/url=12.8552k speed=194.012M/s time/byte=5.15432ns time/url=14.4398us url/s=69.253k/s 

The table-based approach more than 10% faster than our main branch and, according to these tests, faster than PR #473

cc @ttsugriy

This is a precursor to #459