Make csvlint way faster

[jpmckinney]

The speed gain is directly correlated to the size of the CSV. I tested on a 1.7MB CSV. These commits dropped its validation time from 9s to 1s. I used stackprof to identify bottlenecks. Main bottleneck now is Ruby's CSV library, though date format detection is still slow. I recommend reviewing one commit at a time. In brief:

• Using exceptions to control flow is incredibly slow (biggest speed gain here)
• Metaprogramming is slow
• A rescue that swallows all exceptions is an anti-pattern, because it can hide bugs, in this case "TypeError: no implicit conversion of Proc into String"

With the new, fast version, the code spends time roughly as follows:

• 17% garbage collection
• 82% loop in parse_csv
  • 49% csv.shift (previously 13%)
  • 24% build_formats (previously 61%)
  • 8% build_errors
  • 5% col_counts (because Rails' String#blank? is slow)

So, csvlint now spends a lot more time parsing CSVs (a cost we can't reduce) and less time rescuing exceptions (a cost I cut here).

[jpmckinney]

Weird, no test failure locally. I'll try later on Ruby 2.0 (running 2.1). Update: Confirm works in 2.1 but not 2.0. Looking into it.

[coveralls]

Coverage remained the same when pulling e498fe8 on jpmckinney:speed into f926eb5 on theodi:master.

[jpmckinney]

I've got some more commits for more speed, but I don't want to overcomplicate this PR. (e.g. 12MB CSV takes ~10s, not ~230s).

[pezholio]

This is awesome. Thanks so much 😄

I'll review this in more detail and merge tomorrow if all's good 👍

[coveralls]

Coverage remained the same when pulling cf3dc46 on jpmckinney:speed into f926eb5 on theodi:master.

[pezholio]

Hurrah for open source! 🎆