Make value capture of optionals more robust.

[grynspan]

When an expression is optional and is passed through #require(), and that requirement fails, we capture the expression's subexpressions and record them as an expectation-failed event that is eventually propagated to stderr. Yay.

The underlying value-capturing machinery accepts optional values as input so that they can be lazily evaluated: if a value is nil at this layer, that means that it wasn't evaluated and we present "<not evaluated>" instead of a description of it or "nil".

In the case of #require() unwrapping optionals, these values need to be boxed in a second optional (i.e. T?? instead of T?, or Optional<Optional<T>> instead of Optional<T>) so that if the value is nil, it isn't confused with "no value". See the following table:

Expression	Value of x as String??	Presented As
try #require(x)	.some(.some("Hello World"))	"Hello World"
try #require(x)	.some(.none)	nil
try #require("Goodbye Venus" ?? x)	.none	"<not evaluated>"

(This maybe suggests we should use a different enum than Optional here, but that's a debate for another PR I think…)

This PR fixes a couple of spots where we weren't casting our values, known to be optionals, to double-optionals, resulting in incorrect output that incorrectly included "<not evaluated>".

Checklist:

• Code and documentation should follow the style of the Style Guide.
• If public symbols are renamed or modified, DocC references should be updated.

[grynspan]

@swift-ci please test