Comatibility test may trigger segfaults

[IvoDD]

We've seen azure storage compat tests trigger a segfault here:

FAILED tests/compat/arcticdb/test_compatibility.py::test_modify_old_library_option_with_current[4.5.0-azurite] - tests.compat.conftest.ErrorInVenv: Executing ['from arcticdb import Arctic', 'import pandas as pd', 'import numpy as np', "ac = Arctic('azure://DefaultEndpointsProtocol=http;AccountName=devstoreaccount1;AccountKey=Eby8vdM02xNOcqFlqUwJPLlmEtlCDXJ1OUzFT50uSRZ6IFsuFq2UVErCz4I6tq/K1SZFPTOtr/KBHBeksoGMGw==;BlobEndpoint=http://localhost:10047/devstoreaccount1;Container=container3;CA_cert_path=/tmp/tmpos9yrqltAzuriteStorageFixtureFactory/client.pem')", "expected_df = pd.read_parquet('/tmp/tmp2dkdtq4p/expected_df.parquet')", "lib = ac.get_library('test_modify_old_library_option.269_2024-11-04T17_39_52_672424')", "read_df = lib.read('sym').data", 'pd.testing.assert_frame_equal(read_df, expected_df)'] failed with return code -11

We've seen such segfaults previously with lmdb and mongo which both had race conditions at desctruction time. We believe we've fixed the lmdb and mongo segfaults from 4.5.1 going forward.

Due to the nature of the compat tests they are very good at reproducing these type of desctructor segfaults because they run many short lived processes which end with many desctructions.

I suspect we might have a similar desctructor related bug with azure.

[IvoDD]

Also relatedly the mongo segdfault which we taught we fixed in #1862 is still observed in 4.5.1 e.g. here:

FAILED tests/compat/arcticdb/test_compatibility.py::test_modify_old_library_option_with_current[4.5.1-mongo] - tests.compat.conftest.ErrorInVenv: Executing ['from arcticdb import Arctic', 'import pandas as pd', 'import numpy as np', "ac = Arctic('mongodb://localhost:15584')", "expected_df = pd.read_parquet('/tmp/tmpn5c751c7/expected_df.parquet')", "lib = ac.get_library('test_modify_old_library_option.475_2024-11-05T10_46_39_809132')", "read_df = lib.read('sym').data", 'pd.testing.assert_frame_equal(read_df, expected_df)'] failed with return code 139

I had previously identified the segfault is in the mongo destructor.

We need to fix that as well.

[IvoDD]

Aaand lmdb can segfault too:

FAILED tests/compat/arcticdb/test_compatibility.py::test_modify_old_library_option_with_current[4.5.1-lmdb] - tests.compat.conftest.ErrorInVenv: Executing ['from arcticdb import Arctic', 'import pandas as pd', 'import numpy as np', "ac = Arctic('lmdb:///tmp/pytest-of-runner/pytest-0/popen-gw0/test_modify_old_library_option0')", "expected_df = pd.read_parquet('/tmp/tmpzgvmc357/expected_df.parquet')", "lib = ac.get_library('test_modify_old_library_option.966_2024-11-05T13_05_18_218166')", "read_df = lib.read('sym').data", 'pd.testing.assert_frame_equal(read_df, expected_df)'] failed with return code 139

Given it's so widespread this is likely not an issue with our storage destructors. I've recently seen it only on conda linux builds so it might be related to that?