[stdlib] fix UnsafePointer.withMemoryRebound(to:capacity:) argument t…

[atrick]

SE-0107 states that UnsafePointer.withMemoryRebound(to:capacity:) should produce
a const UnsafePointer, but the implementation that I committed in Whitney
produces an UnsafeMutablePointer.

As a result Swift 3 accepts code, that we would like to reject:

func takesUInt(: UnsafeMutablePointer) {}
func takesConstUInt(: UnsafePointer) {}

func foo(p: UnsafePointer) {
p.withMemoryRebound(to: UInt.self, capacity: 1) {
takesUInt($0) // <========= implicitly converts to a mutable pointer
takesConstUInt($0)
}
}

We would like to reject this in favor of:

func takesUInt(: UnsafeMutablePointer) {}
func takesConstUInt(: UnsafePointer) {}

func foo(p: UnsafePointer) {
p.withMemoryRebound(to: UInt.self, capacity: 1) {
takesUInt(UnsafeMutablePointer(mutating: $0))
takesConstUInt($0)
}
}

This looks to me like an experimental change accidentally creeped onto my branch
and it was hard to spot in .gyb code. I needed to write the unit test
in terms of UnsafeMutablePointer in order to use expectType, so the original test didn't
catch this.

rdar://28409842 UnsafePointer.withMemoryRebound(to:capacity:) incorrectly produces a mutable pointer argument

[atrick]

CCC Info:
• Explanation: UnsafePointer implementation in Swift 3 does not match the proposal.
• Scope of Issue: The Swift 3 API is too lenient in one case. Rarely, users who already migrated to 3 may need to insert an additional cast after this fix.
• Origination: SE-0107: Swift 3.
• Risk: Potentially causes a compiler error in code that was already migrated.
• Nominated Reviewers: @dabrahams
• Testing: Unit test. Currently testing build-toolchain on swift-3.0-branch.
• Bug: rdar://28409842

[atrick]

@swift-ci test.

[dabrahams]

LGTM!