Incomplete responses lead to failure-loop

[Joacchim]

With the fix for #54 , a new bug was introduced:
When the reponse cannot be read in its entirety (whatever the reason, but the bug was shown thanks to an error in the DELETE op of the playground server), the sock_send_receive function tries to re-send the request, which was partially overwritten while reading the response.

This means that either:

• We must "receive" in a different temporary buffer
• We must not retry to transmit the request

Also:

• The retry must be limited to a given number of times (otherwise the fail loop won't ever be stopped)
• The status returned when the fail-loop is interrupted or when a partial response was received should reflect the fact that the response isn't complete (returning a negative status ? Which errno code? another method ?)

[Joacchim]

Since a few retries mechanisms are implemented on top of those functions, I'm guessing that it's not that useful to have it within the function (or we should completely do it within the function, and not on top of it at all ?)

So my plan is to remove the retry when some data could be read (maybe even in all cases, I'm not sure ?), AND to return an errno code (probably EIO). What do you think about it ?

[Joacchim]

Discussed with vianney, should be ok with #60 as an addition.

[Joacchim]

Cannot reproduce it anymore with the playground server, I personally consider this as fixed; although it might be better to test a bit more.
Previous failing scenario:

• Send a request through the driver
• response partially sent
• the CDMI server closes the connection, provoking a retry on the SRB side
• the buffer of the request/responses now is partially erased by the response
• The CDMI server returns a 400 Bad Request because the request has been corrupted when reading the beginning of the response.