The curl file adaptor uses libcurl to perform remote  file transfer copies.
Other methods than file.copy are not implemented at the moment.

The curl_context adaptor provides means to specify SSH, HTTP, HTTP_POST and
other security credentials (Note: net yet implemented)

Note to self: we will now have two ssh contexts - how can we distingush them?
We probably should not have two in the first place, but if only one suite is
build, the context is not available for the other...   *FIXME*

Each SAGA file copy operation results in two curl file copies, to facilitate
third party file transfer transparently.  The first operation streams the src
file to a read callback function.  That function puts the data into a buffer.
That buffer is read by the second operation, i.e. by that operations write
callback function.  Independent curl handles are used for these operations.

Note that at the moment, we use  curls internal callback functions to buffer the
data in a file - the respective FILE* handles are created in the adaptor.

Curl is thread save, however operations on the same curl handle are not.  The
adaptor is thus maintaining a registry of curl handles per thread.