Improve iobuf -> numpy conversion latency

Summary: roticv provided this change to bypass coalescing iobuf with `b"".join(iobuf)` and providing the iobuf itself as the buffer to numpy array

Reviewed By: roticv

Differential Revision: D69883091

fbshipit-source-id: 64f6268237f2e8de2b221ba3e1d0bf068c8e5f28

folly/python/iobuf.pxd

@@ -32,6 +32,7 @@ cdef extern from "folly/io/IOBuf.h" namespace "folly":
         size_t countChainElements()
         uint64_t computeChainDataLength()
         unique_ptr[cIOBuf] clone()
+        unique_ptr[cIOBuf] cloneCoalesced()
         cIOBuf* prev()
         cIOBuf* next()
         void insertAfterThisOne(unique_ptr[cIOBuf]&& ciobuf)

folly/python/iobuf.pyi

@@ -18,6 +18,7 @@ class IOBuf(Hashable):
     def __init__(self, buffer: Union[IOBuf, bytes, bytearray, memoryview]) -> None: ...
     def writable(self) -> bool: ...
     def clone(self) -> IOBuf: ...
+    def cloneCoalesced(self) -> IOBuf: ...
     @property
     def next(self) -> Optional[IOBuf]: ...
     @property

folly/python/iobuf.pyx

@@ -103,6 +103,10 @@ cdef class IOBuf:
         """ Clone the iobuf chain """
         return from_unique_ptr(self._this.clone())
 
+    def cloneCoalesced(self):
+        """ Clone the iobuf chain """
+        return from_unique_ptr(self._this.cloneCoalesced())
+
     @property
     def next(self):
         _next = self._this.next()