Performance improvements

[cheald]

After running ruby-prof around bundler a few times, I noticed that it was spending a lot of time in LazySpecification#== and LockfileParser#parse_spec. This patch is just a simple refactoring of those two sticky spots to be faster. Additionally, I moved constant strings and regexes into constants to avoid extraneous per-line garbage generation when parsing a lockfile.

• LazySpecification now has an #identifier method that is a memoized hash value of the name, version, source, platform, and dependencies. This gives us a quick way to effectively test eql?, which has implications in various places.
• LazySpecification#== now uses #identifier rather than building and doing array comparisons.
• LockfileParser accumulates specs into a hash keyed by #identifier rather than testing @specs.include? for every spec. This reduces the runtime from O(n^2) to O(n log n).

All specs pass under Ruby 1.9.3-p327.

To demonstrate the impact, here's my before-and-after times for a production bundle:

                Before (sec)   After (sec)   % of original
bundle install  1.22           0.89          72.95%
bundle check    0.57           0.31          54.39%
bundle list     0.65           0.38          58.46%

[indirect]

Super awesome. Thanks!