fix: Deepcopy flag buffer on scope fork

[cmanallen]

A previous attempt to fix this issue focused on a related issue and assumed that it would fix this. This PR adds a test using the reproduction steps mentioned in the bug report and fixes the issue by deep copying the buffer.

Closes: #3852

[codecov]

❌ 1 Tests Failed:

Tests completed	Failed	Passed	Skipped
13814	1	13813	4139

View the top 1 failed tests by shortest run time

tests.integrations.stdlib.test_httplib test_http_timeout

Stack Traces | 75.2s run time

.../integrations/stdlib/test_httplib.py:373: in test_http_timeout
    assert len(transaction["spans"]) == 1
E   assert 0 == 1
E    +  where 0 = len([])

To view more test analytics, go to the Test Analytics Dashboard
📢 Thoughts on this report? Let us know!

[cmanallen]

@antonpirker Re-ran the test and this is now passing. Has test_http_timeout always been flakey? Maybe it should be skipped?

[ffelixg]

If you remove this method entirely, aren't we back at square one? I.e. the cache will get polluted by changes to a copy:

def test_mutate():
    cache = LRUCache(1)
    cache.set(0, 0)
    copy(cache).set(0, 1)
    assert cache.get(0) == 0 # fails

[cmanallen]

No because the call site is now using deepcopy. Which targets the dunder deepcopy method which we leave as default.

[antonpirker]

Closing this in favor of solution linked here: #3852