lock SendData to fix random connection failures

[mus65]

Fixes #1253 and probably #1493 .

This adds a lock to SendData to avoid the loop to interleave between multiple messages, causing a "bad message" error in the server (see #1253 (comment)).

This may totally be the wrong fix since I'm not really familiar with the protocol and I couldn't find anything specifically about this in the RFCs. But it seems sensible to me that the server doesn't expect this to happen.

By adding some logging I confirmed that in the failure cases of the test, the loop did indeed interleave between different messages. In the success cases, there simply were never more than 1 iteration, so it couldn't interleave.

Without this fix, I had the test fail about every third run. With this fix, I have run the test for about 500 times without a failure.

[Rob-Hague]

Nice! It makes sense: the SFTP layer has its own length-prefixed packets so when they are being (wrongly) split up in the channel layer, the server reads an SFTP packet, tries to read the length of the next SFTP packet but ends up reading some random bytes in the middle of a different packet. Good find