Load data with Marshal instead of YAML

[markprzepiora]

Loading 4MB of YAML data takes about 1 second on my development machine. Replacing the YAML-dumped data with Marshal-dumped data instead brings this down to about 0.08 seconds. This makes a big difference in feedback speed especially when running individual tests that rely on ZipCode.identify.

                      user     system      total        real
YAML.load         0.859559   0.029983   0.889542 (  0.889713)
Marshal.load      0.043685   0.030232   0.073917 (  0.073922)

[Hengjie]

This looks awesome. Thank you for writing it so that it can be sped up. I just wish the gem author merged it now.

[brodyhoskins]

@markprzepiora, I'd like to explore options for storing the data.

What I ended up doing in a fork was converting the YAML to CSV and using the FastCSV gem to process the data more quickly and to prevent loading it into memory all at once; however I don't feel that it was necessarily the best method.

Another idea is to keep the YAML around for development purposes but bundle a SQLite database generated by a rake task.

Since you've opened this PR I wanted to get your feedback.

[NewAlexandria]

This is a breaking change (semver-1)

Suggested change

	VERSION = '0.2.1'
	VERSION = '1.0'

If it's going to be included one day, I think the PR needs to include a build script, that convert a YAML (easy for humans to deal with) into the marshaled format.

[NewAlexandria]

Feel free to also open a PR against https://github.com/suvie-eng/zip-codes/

I'm not wanting to create maintainer controversy - but until the maintainers here integrate all of these, I'd like to have a gem that is the most up to date. Also, I like the performance of this change 👌

[lostapathy]

Library data like this really shouldn't be distributed via Marshal - the format of Marshal is not guaranteed to be stable between ruby versions (and indeed, is not in practice). I'd suggest either pursuing fast CSV options or just querying a sqlite file.