Issue Description:

We should fix both UnsafePointer and UnsafeBufferPointer to handle converting to a type with a different stride.

For UnsafePointer, document that the capacity argument pertains to the number of elements of the to type (not the from type).

For UnsafeBufferPointer, the capacity is derived from the buffer capacity.

This can be done by adding checks to the stdlib implementation:

First check if the to/from strides are equal. If so, no additional code should be executed as a result of this change.

Next find the stride ratio. Either the from or to type may be larger than the other, but the larger stride must be an exact multiple of the smaller.

For UnsafeBufferPointer, call the first bindMemory with the adjusted capacity. For UnsafePointer the second bindMemory call's capacity changes.

If the to type is larger, add an assertion that the buffer is aligned to to.align.

This merits a discussion on the swift-evolution list. But since it's only broadening the accepted input, I don't think it needs an SE proposal. The documentation does need to change.

This has been discussed before. I balked at supporting this right away because I wanted a zero-abstraction cost guarantee for all new memory binding APIs (relative to simple pointer conversion). But, given the code that I've seen since then, I think it's worth doing and just hoping compile always specializes the call.