Optimize "uniq" to be O(n) instead of O(n^2)

[Tzook]

The uniq function does heavy operations that could cause performance issues on big arrays.
This commit utilizes the Map object if supported (95% browser support according to caniuse), to make such cases much faster.
Running a simple example in Chrome on an array with 10,000 unique elements:

• Before - takes about 8 seconds
• After - takes about 0.03 seconds

[coveralls]

Coverage increased (+0.02%) to 96.887% when pulling e462042 on Tzook:master into 53086b0 on jashkenas:master.

[jdalton]

Keep in mind this toggles the matching behavior from strict equality to same value zero depending on environment support.

[Tzook]

Good point @jdalton.
The only differences according to this table are that -0 is not equal to +0, and that NaN equals to NaN.
In my opinion it makes even more sense for the function to use same-value-zero comparison. Having an array with multiple NaNs should result in only one NaN after using unique.

[jdalton]

Yep.

[forty]

Hello everyone,
is this PR ever going to be merged, or does it need additional work?

Thanks

[jgonggrijp]

• It would have to consistently use SameValueZero, even in environments that don't have Map.
• There would have to be a new major release of Underscore in which we actually decided to use SameValueZero instead of strict equality everywhere. See Change to SameValueZero for comparisons #2453. It is not yet decided whether we will ever do that.
• It shouldn't do if/else all over the place. Move the supportsMap out of the function and switch over its value only once instead of repeatedly in every loop iteration.