`guard` capture specifier for closure capture lists

[xwu (Xiaodi Wu)]

@cal

If [assert weak self] existed, I suspect it would become the status quo default for Void closures, since it has better ergonomics (it avoids some boilerplate and enabled implicit self). This would mean the status quo in this case would no longer be silently ignoring potential failures, which would be unambiguously better with respect to correctness.

Other than differences in the diagnostics (which, granted, is no small thing), what is the difference between this and [unowned self] without the flexibility of choosing where to assert?