fix(merge-rules): don't change specificity of prefixed properties

[clshortfuse]

fixes #717

[coveralls]

Coverage decreased (-0.03%) to 99.104% when pulling 535933f on clshortfuse:fix/mergerules-avoidvendoroverride into cdedda7 on cssnano:master.

[clshortfuse]

This is a pretty significant change. It essentially a rewrite of the merging function. The logic is mostly the same save for some optimizations.

• Instead of converting the postCss.Rule nodes to string, the code now directly uses the postCss.Declaration class. It helps a bit with type-checking and ensure we're using the right objects. It also avoids a couple of string concatenation and manipulation functions.
• Instead of iterating through all of the declarations in the first rule in reverse, we iterate through the collected set of intersections. Each intersected declaration is then compared against the first rule's declarations and any rule with a later override/conflict (based on property name) is filtered out.
• After, instead of iterating through all declarations inside second rule, we iterate through the previously filtered intersections. For each intersected declaration, we filter, in order, those that do not appear in the same order as in the second rule's declarations, based on the declaration's base property name (without prefixes).

This allows use to safely merge new rulesets, such as:

.a{
  transform:scaleY(1);
  -webkit-transform:scaleX(-1)
}
.b{
  -webkit-transform:scaleX(-1);
  transform:scaleY(1)
}

Here, -webkit-transform:scaleX(-1) could safely be merged together, despite both having other transform rules.

Also, we can now merge this:

  .a{ display:block }
  .b{ display:block }
  .c{ display:block; display:inline-block}

Previously, the algorithm wouldn't break apart the display rules in .c because it wasn't analyzing the rule order. Now, the result would be: .a,.b,.c{display:block} .c{display:inline-block};

[alexander-akait]

Good job, thanks for helping, maybe you can help with other issues? Just don't have time on this right now, anyway thanks again