End-to-end tests are unstable on CI

[jleibs]

The intel-mac e2e tests were failing and in an attempt to unblock our wheel builds, I disabled them in #1853

In the commit before disabling the mac builds, the linux e2e were passing. You can see an example successful run here:

• 3cd534f
• https://github.com/rerun-io/rerun/actions/runs/4701577971/jobs/8337781980

However after landing the change to disable the tests, the linux tests are now failing as well:

• 5dc9b61
• https://github.com/rerun-io/rerun/actions/runs/4702053139/jobs/8338847034

Running the same commit locally runs fine.

The stack trace here hints at a shutdown race condition, I think related to the multi-processing test.

[2023-04-14T17:59:28Z INFO  rerun::run] Received first message.
31
thread 'msg_encoder' panicked at 'Main thread should still be alive: "SendError(..)"', crates/re_sdk_comms/src/buffered_client.rs:203:47
32
stack backtrace:
33
[2023-04-14T17:59:34Z INFO  rerun::run] Successfully ingested 20002 messages.
34
   0: rust_begin_unwind
35
             at /rustc/d5a82bbd26e1ad8b7401f6a718a9c57c96905483/library/std/src/panicking.rs:575:5
36
   1: core::panicking::panic_fmt
37
             at /rustc/d5a82bbd26e1ad8b7401f6a718a9c57c96905483/library/core/src/panicking.rs:64:14
38
   2: core::result::unwrap_failed
39
             at /rustc/d5a82bbd26e1ad8b7401f6a718a9c57c96905483/library/core/src/result.rs:1791:5
40
   3: re_sdk_comms::buffered_client::msg_encode

[jleibs]

If there's a regression here (which is unclear since I can't repro locally) this might be a candidate worth looking at: #1830

[jleibs]

This workaround has at least hidden the issue: #1857

But I think we should leave this open until we've investigated a bit further. The fact that this did not produce the error that would have been expected is additionally concerning.