Code adjustments in unique() generator

[vlakoff]

Highlights:

• When unique value found, serialize once instead of twice. If unique values are likely to be found on first try, this results in twice less serialize calls. Refs Serialized $res key #749.
• Use isset() instead of array_key_exists(), the former being a language construct and not a function, it is significantly faster. This change also allows code simplification by removing the "intermediary index creation" step at beginning of method. Refs Improve unique generator performance #491.

[fzaninotto]

Did you benchmark the speedup?

[vlakoff]

Just did some quick xdebug profiling, and the speedup about serialize() and array_key_exists() appears to be rather insignificant… (still, it can't harm, let's call this a little code polishing ;)

Though, the call_user_func_array() calls are very costly, they are a bottleneck (roughly 10% execution time of my test script, top 5 of most costly functions). They can be avoided as Laravel does, and I think the added bit of complexity is worth it.

After this PR is merged (if you wish so ;) I'll submit an other one to avoid calling this hugely costly call_user_func_array().

[fzaninotto]

If you submit a perf patch, you must be able to prove that it works. So please provide an execution time before/after on a series of calls, describing the conditions of the test. If the gain is significant, I'll merge. Otherwise, I won't. It's as simple as that.

[vlakoff]

Please consider this PR as just some code adjustments, merge it or not, up to you :)

If I submit a PR about call_user_func_array(), I'll provide some numbers along.

[vlakoff]

Rebased and simplified a little bit.

Despite the diffs, the changes are really straightforward. It just brings a cleaner code.

Could you please merge or close? your pick :)