Rounding of output numbers.

[seven-phases-max]

Fixes #1326, #1810, #1524.

This probably needs some discussion/changes before merging: I hardcoded precision to max. 8 digits after the decimal point but did that via "environment constant" so it can be exposed as a compiler option if necessary (on the other hand if we don't need such option this code may be optimized for better performance.. but that's to be done with some care and more extensive tests since the same code is used in internal calculations for pattern-matching and guard-expressions: see "known issue 2" below).

Known issues:

(1). Rounding is not applied to "hand-written" numbers immediately assigned to properties, e.g:

z {
    @value: 0.999999999;

    1: @value;            // -> 1
    2: 0.999999999 /* */; // -> 1
    // but:
    3: 0.999999999;       // -> 0.999999999
}

There the third value is of tree.Anonymous type which is not rounded. It is possible to round tree.Anonymous too (when it's actually numeric), but I believe it's better to be fixed on another level (not a critical issue anyway since it's hard to imagine why you would need to type 0.999999999 and then want it to be rounded to 1).

(2). Rounding is not applied to values used in "internal" number evaluations like pattern-matching and guard-expressions, e.g.:

z {
    @value: alpha(shade(#000, -15%));
    1: @value;   // -> 1;
    .m1(@value); // RuntimeError: No matching definition was found for `.m1(0.9999999999999999)`
    .m2(@value); // does not match anything for the same reason (@value = 0.9999999999999999)
}

.m1(1) {
    m1-matched: true;
}

.m2(@value) when (@value = 1) {
    m2-matched: true;
}

This is possible to improve/fix using the same code of this PR, but needs some changes in "environment" options passing/handling (and there's some env trickery, so I did not touch it just to not break something critical). So there's room for discussion and further improvements.

Aside from above everything works as expected (see changes in tests).

[matthew-dean]

Nice work. #2 I agree is a problem, but seems like a step in the right direction.

[lukeapage]

Looks good.

re: 1, I'd like to actually change that place in the parser that sees if the value is simple enough to make it anonymous to always try and parse

[lukeapage]

p.s. related issue on clean-css
clean-css/clean-css#163