Use tobytes in hex to make data C contiguous

Only copy the data in `hex` using `tobytes` if it is not C contiguous.
That way if the data is already C contiguous, a view onto the original
data can be used instead saving time and memory. The copy is needed for
non-C contiguous data to get a consistent representation in hex as
`memoryview` and other objects do. As Python 2's `binascii.hexlify`
supports the (new) buffer protocol, it won't introduce any additional
copies as long as something implementing the (new) buffer protocol is
provided. On Python 3, we leverage `memoryview`'s `hex` method, which
also avoids copying if the data is C contiguous. It provides a fast path
to `hex`, which is equivalent to the `bytes` method `hex`. Thus in both
cases, the hex conversion will avoid any additional copies of the data.
So this should in the best case avoid a copy when constructing the hex
result, but may still copy the data if it is not contiguous in the way
we expect, which is better than before.

src/cybuffer.pyx

@@ -35,6 +35,8 @@ include "version.pxi"
 cdef extern from "Python.h":
     size_t Py_UNICODE_SIZE
 
+    object PyMemoryView_FromObject(object obj)
+
 
 cdef extern from *:
     """
@@ -269,11 +271,18 @@ cdef class cybuffer(object):
 
 
     cpdef str hex(self):
+        cdef object d
         cdef str s
+
+        if self.c_contiguous:
+            d = self
+        else:
+            d = self.tobytes()
+
         if PY2K:
-            s = hexlify(self.tobytes())
+            s = hexlify(d)
         else:
-            s = self.tobytes().hex()
+            s = PyMemoryView_FromObject(d).hex()
 
         return s