Support spread operator in call expressions

[ahejlsberg]

This PR adds support for the spread (...) operator in call expressions with down-level code generation for ES3/5. For example:

function drawText(x: number, y: number, ...strings: string[]) {
    // draw text
}

drawText(10, 20, "hello");
drawText(10, 20, "hello", "world");
var a = ["one", "two"];
drawText(10, 20, ...a);
drawText(10, 20, "zero", ...a, "three");

The function calls are emitted unchanged for ES6, but when the target is ES3 or ES5, the calls are rewritten as follows:

drawText(10, 20, "hello");
drawText(10, 20, "hello", "world");
var a = ["one", "two"];
drawText.apply(void 0, [10, 20].concat(a));
drawText.apply(void 0, [10, 20, "zero"].concat(a, ["three"]));

Type checking requires spread elements to match up with a rest parameter. Thus, in a typed function call it is not possible to use a spread element for non-rest parameters. For example:

var pos = [10, 20];
drawText(...pos, "hello", "world");  // Error
(<any>drawText)(...pos, "hello", "world");  // But ok in untyped call

A temporary variable may be injected to ensure the this argument is only evaluated once. For example:

interface Context {
    drawText(x: number, y: number, ...strings: string[]);
}

declare function getContext(): Context;

var text = ["hello", "world"];
getContext().drawText(10, 20, ...text);

This generates the code:

var text = ["hello", "world"];
(_a = getContext()).drawText.apply(_a, [10, 20].concat(text));
var _a;

[yuit]

This class C/D can be pull out into its own test cases