Attachment lifetimes

[grynspan]

This PR introduces a new experimental trait, .savingAttachments(if:), that can be used to control whether a test's attachments are saved or not.

XCTest has API around the XCTAttachment.Lifetime enumeration that developers can use to control whether attachments are saved to a test report in Xcode. This enumeration has two cases:

/*
 * Attachment will be kept regardless of the outcome of the test.
 */
XCTAttachmentLifetimeKeepAlways = 0,

/*
 * Attachment will only be kept when the test fails, and deleted otherwise.
 */
XCTAttachmentLifetimeDeleteOnSuccess = 1

I've opted to implement something a bit more granular. A developer can specify .savingAttachments(if: .testFails) and .savingAttachments(if: .testPasses) or can call some custom function of their own design like runningInCI or hasPlentyOfFloppyDiskSpace. The default behaviour if this trait is not used is to always save attachments, which is equivalent to XCTAttachmentLifetimeKeepAlways. XCTAttachmentLifetimeDeleteOnSuccess is, in effect, equivalent to .savingAttachments(if: .testFails), but I hope reads a bit more clearly in context.

Here's a usage example:

@Test(.savingAttachments(if: .testFails))
func `best test ever`() {
  Attachment.record("...") // only saves to the test report or to disk if the
                           // next line is uncommented.
  // Issue.record("sadness")
}

I've taken the opportunity to update existing documentation for Attachment and Attachable to try to use more consistent language: a test records an attachment and then the testing library saves it (somewhere). I'm sure I've missed some spots, so please point them out if you see them.

Resolves rdar://138921461.

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]

This code isn't new, but needed to move from handleEvent as it imposed the wrong layering (saving the attachment before any event handlers actually ran.)