Fixed diffie-hellman shared key computation

[pmconrad]

The shared secret computed by DH_compute_key can be smaller than the estimate provided with DH_size. This must be accounted for, otherwise there will be uninitialized trailing bytes in the resulting shared_key vector.
https://www.openssl.org/docs/man1.0.2/crypto/DH_compute_key.html

[jmjatlanta]

Suggestion: The test must be run several times for keys to be generated that produce a shared key that is smaller than the estimated size. Perhaps adjust the test to run several iterations (15-25 in my testing), or hard-code the user keys so that they always produce a shared key that is smaller.

[jmjatlanta]

Works well.

[pmconrad]

Hm, travis failed with this (unrelated) error:

1306549ms th_a task_cancel.cpp:245 test_method ] 9 canceled_exception: Canceled
task simple_task canceled before starting, reason canceling scheduled task to test if cancel works
{"description":"simple_task","reason":"canceling scheduled task to test if cancel works"}
th_a task.cpp:50 run_impl
/home/travis/build/bitshares/bitshares-fc/tests/thread/thread_tests.cpp(65): error in "yields_execution": check "hello world" == result failed [hello world != worldhello ]

Looks like a race condition somewhere. Hopefully in the test only. :-/

[pmconrad]

There is a race condition in thread_tests/yields_execution: if the first task executes its yield() before the second task has been created, the yield() is a no-op and "world" is written to the result before "hello ". The race condition can be reproduced by inserting a usleep() between the creation of the two async tasks.

I have tried fixing this with a spin lock, but didn't succeed. In any case this is unrelated to this PR, so I'm merging now.