Be a bit more frugal with our compression buffer

For small files (and the remaining bit of large ones, though this is not
as big of a deal), the compression buffer can be sized based on the
remaining data rather than the block size. This often allows it to avoid
being classified as a "medium sized" allocation, saving an expensive
call to madvise when the buffer is freed.

unxip.swift

@@ -127,8 +127,9 @@ struct File {
  compressionStream.addTask {
  try Task.checkCancellation()
  let position = _position
- let data = [UInt8](unsafeUninitializedCapacity: blockSize + blockSize / 16) { buffer, count in
- data[position..<min(position + blockSize, data.endIndex)].withUnsafeBufferPointer { data in
+ let end = min(position + blockSize, data.endIndex)
+ let data = [UInt8](unsafeUninitializedCapacity: (end - position) + (end - position) / 16) { buffer, count in
+ data[position..<end].withUnsafeBufferPointer { data in
  count = compression_encode_buffer(buffer.baseAddress!, buffer.count, data.baseAddress!, data.count, nil, COMPRESSION_LZFSE)
  guard count < buffer.count else {
  count = 0