Multiple instances of RemoteAgentBuffer will cause seg fault

[Gamenot]

Description

Instantiating multiple instances of RemoteAgentBuffer will cause a segmentation fault. This is a problem because it blocks having more than one instance of SMARTS from being instantiated in the same process.

Context

I have tested to see that you can have two instances of the Renderer in the same thread so I think this PR accomplishes the intention. I still would prefer to see a simple test for this.

I also understand why there is a rendering lock... I wonder about the implications of it on performance but it is enough that rendering for multiple simulations is possible on the same process.

In addition, what I have found is that the RemoteAgentBuffer currently prevents two instances of SMARTS from existing on the same thread so I am adding that as an issue.

Originally posted by @Gamenot in #706 (comment)

[sah-huawei]

I've confirmed this is the case and is related to the version of the grpcio dependency.
Version 1.31.0 causes a segfault in the simple test in PR #745 on this line grpc.channel_ready_future(self._worker_channel).result(timeout=30) in the RemoteAgent __init__ function.
(Using gdb, it can be seen the segfault is caused by a Flush in the grpc C code. Probably a threading bug.)

Versions 1.35.0 and 1.37.0 (the latest) don't segfault.

[sah-huawei]

@Adaickalavan

[sah-huawei]

This is currently the only (remaining) reason we can't run multiple instance of SMARTS in the same process, and so also can't run our tests without --forked cf. Issue #743.