Add search index

[apainintheneck]

Resolves #14.

Overview

The main change here is to pre-compute all word to country mappings for performance reasons. This adds quite a bit of complexity to the code though because we have to make sure that these mappings stay up-to-date.

Changes

• Add script/shared/cache_generator.rb class to make it easier to generate cache files with documentation.
• Add script/generate_search_index.rb to facilitate generating cached search indexes and checking to make sure they are up-to-date.
• Add cache.rb module to facilitate getting values from the cache.
• Update code to use new search indexes.
• Refactor: Move generate_readme.rb to new script/ folder.
• Add rake task to generate and check for outdated search indexes.
• Add non-readme cache files to the gem files (update in atlasq.gemspec).
• Check for outdated indexes on CI.
• Add info about the cache to the readme in the dev section.

Benchmarks

The performance gains come from two main areas.

1. We have mappings country names and country name fragments which are pre-computed so we don't have to allocate 100s of thousands of strings when we normalize strings at runtime.
2. We no longer need to load the internationalization library for 93 different languages. We can just use the default English one.

This even notably speeds up tests from around 8 seconds to around 4 seconds.

---

This is a measurement of the worst-case scenario before this change since it required creating a partial country name index on the fly.

$ hyperfine \
  --parameter-list branch main,add-search-index \
  --setup 'git switch {branch}' \
  --warmup 3 \
  'exe/atlasq -c united'

Benchmark 1: exe/atlasq -c united (branch = main)
  Time (mean ± σ):      1.211 s ±  0.019 s    [User: 0.899 s, System: 0.275 s]
  Range (min … max):    1.191 s …  1.256 s    10 runs
 
Benchmark 2: exe/atlasq -c united (branch = add-search-index)
  Time (mean ± σ):     621.9 ms ±   3.2 ms    [User: 341.8 ms, System: 252.7 ms]
  Range (min … max):   617.8 ms … 628.6 ms    10 runs
 
Summary
  exe/atlasq -c united (branch = add-search-index) ran
    1.95 ± 0.03 times faster than exe/atlasq -c united (branch = main)

The results would look like this.

$ exe/atlasq -c united
*
* Countries (Partial Match)
* * * * * * * * * * * * * * *
(🇦🇪 | 784 | AE | ARE | United Arab Emirates)
(🇬🇧 | 826 | GB | GBR | United Kingdom of Great Britain and Northern Ireland)
(🇲🇽 | 484 | MX | MEX | Mexico)
(🇹🇿 | 834 | TZ | TZA | Tanzania, United Republic of)
(🇺🇲 | 581 | UM | UMI | United States Minor Outlying Islands)
(🇺🇸 | 840 | US | USA | United States of America)
(🇻🇮 | 850 | VI | VIR | Virgin Islands (U.S.))

---

For normal country matching the results are still significant.

$ hyperfine \
  --parameter-list branch main,add-search-index \
  --setup 'git switch {branch}' \
  --warmup 3 \
  'exe/atlasq -c france'

Benchmark 1: exe/atlasq -c france (branch = main)
  Time (mean ± σ):     895.3 ms ±  10.6 ms    [User: 594.2 ms, System: 266.4 ms]
  Range (min … max):   874.3 ms … 908.2 ms    10 runs
 
Benchmark 2: exe/atlasq -c france (branch = add-search-index)
  Time (mean ± σ):     592.8 ms ±   4.4 ms    [User: 318.2 ms, System: 248.4 ms]
  Range (min … max):   585.8 ms … 598.8 ms    10 runs
 
Summary
  exe/atlasq -c france (branch = add-search-index) ran
    1.51 ± 0.02 times faster than exe/atlasq -c france (branch = main)

The results should look like this.

$ exe/atlasq -c france
*
* Country: The French Republic
* * * * * * * * * * * * * * * * *
(🇫🇷 | 250 | FR | FRA | France)
 | Search Term: france
  | Languages: French
   | Nationality: French
    | Region: Western Europe
     | Continent: Europe
      | Currency: € Euro
       |________________________________________

---

Next Steps

It would be nice to not have to bundle all of the data dependencies with the program. These could easily be added to the cache reducing the number of runtime dependencies.