[Static if] replace overload constraints with static if (mutation.d)

[wilzbach]

Split up of #5145

[wilzbach]

This contains a fix for @WalterBright's favorite example of the remove overloads.

[wilzbach]

The next overload which is grouped, is also an InputRange, but also an OutputRange

[wilzbach]

The next overload is is(Range == char[]) || is(Range == wchar[]) which can be appended here as && has a higher precedence.

[wilzbach]

This is @WalterBright's favorite example. There is absolutely no need to expose different overloads depending on the SwapStrategy to the user.

[wilzbach]

The next overload is isRandomAccessRange!Range && hasLength!Range which implies bidirectionality of Range as isRandomAccessRange requires the following:

static assert(isBidirectionalRange!R || isForwardRange!R && isInfinite!R);

Hence it either needs to have swappableElements (current) or is RandomAccess (next)

[wilzbach]

As suggested this puts the overloads into one function, s.t. documentation & error message is improved

[JackStouffer]

Nitpick, change to SwapStrategy.semistable is not supported

[JackStouffer]

Looks good otherwise

[wilzbach]

Looks good otherwise

Addressed the nit.

[MartinNowak]

This broke the nightly builds @wilzbach, see http://nightlies.dlang.org/dmd-master-2017-05-13/build.html.
Please either fixup or revert this PR.

I'm also not positive this is a good long-term direction in general, hope we can work on improved error messages for template constraints later this year.

[JackStouffer]

@MartinNowak The motivation is documentation simplification and not error messages.

[MartinNowak]

https://dlang.org/phobos/std_algorithm_mutation.html#copy
I see, but now it no longer mentions the constraints/required concepts at all.
Of course a private and undocumentd areCopyCompatibleArrays constraints is an anti-pattern as well.

[MartinNowak]

Reduced test-case:

import std.file, std.algorithm;

void main()
{
    string a, b;
    copy(a, b);
}

This actually seems to break "instantiation failure is not an error", for the copy overloadset, because the constraints are too general, now both imported copy functions are callable, but one already errors during semantic (think within the overload resolution).
Please have a look at preserving the existing constraints.