Place RTI OpenSSL on the (LD_LIBRARY_)PATH on Linux

[mikaelarguedas]

This implements "option a" described in ros2/ros2#481 (comment) as option b was evaluated as not acceptable for users.

This also works on xenial as the env vars stay the same so it can be merged async from ros2/ci#148 (but there's not much incentive to merge it before)

Xenial job:
Bionic job:

connects to ros2/ros2#481

[mikaelarguedas]

This is broken on windows, moving back to "in progress" for now

• Linux
• Linux-aarch64
• macOS
• Windows (failure unrelated)

Bionic: (failure unrelated ros2/rclcpp#470)

Rebased and modify logic to pass the build on windows

[mikaelarguedas]

I changed this to modify the PATH only when testing Connext on Linux to make sure we still exercise OpenSSL 1.1 when using Fast-RTPS.
This doesn't modify the PATH for Windows and MacOS but we could use RTI's OpenSSL libraries on these platforms by just removing this condition. As there is no need for now I didnt do it.
If we were to test RTI's OpenSSL on other platforms we would also need to modify this logic to modify the *LIBRARY_PATH on all platforms to include RTI's OpenSSL libraries.

CI with last changes:

• Linux

• Linux-aarch64

• macOS

• Windows

• Bionic

[dirk-thomas]

Reading the patch I was confused about TEST_PATH_WITH_RTI_BIN assuming it would contain the extra bin for all rmw. Further down I see that it is only populated in the case of Connext. Maybe renaming the variable to something like TEST_PATH would avoid that confusion.

Anyway the patch looks fine to me.

[mikaelarguedas]

Reading the patch I was confused about TEST_PATH_WITH_RTI_BIN assuming it would contain the extra bin for all rmw. Further down I see that it is only populated in the case of Connext. Maybe renaming the variable to something like TEST_PATH would avoid that confusion.

Good point I didnt rename the variable after changing the logic: done in 6dbd610