Sampling context improvements

[sentrivana]

• make serialization of args, kwargs consistent
  When using _serialize_span_attribute, it's not guaranteed that the same kind of object will be serialized the same way. E.g. lists are sometimes serialized as lists (if they only contain one element type) but can also be serialized as strings. This makes for a really bad user interface. So just use str everywhere when dealing with serializing task queue args and kwargs.
• unfurl args, kwargs
• better wording in migration guide
• better error handling
  Sometimes we were just silencing errors with try...except. In cases where this can legitimately be an SDK issue, use capture_internal_exceptions instead.
• add missing type hints
• add more span attributes
  GH research shows some folks use headers to make sampling decisions. Add headers to span attributes.

Closes #3746

[codecov]

❌ 305 Tests Failed:

Tests completed	Failed	Passed	Skipped
19178	305	18873	4435

View the top 3 failed tests by shortest run time

tests.test_scope test_get_current_scope_tags

Stack Traces | 0.001s run time

tests/test_scope.py:341: in test_get_current_scope_tags
    assert scope1.client.__class__ == NonRecordingClient
E   AttributeError: client

tests.test_scope test_get_current_scope_tags

Stack Traces | 0.001s run time

tests/test_scope.py:341: in test_get_current_scope_tags
    assert scope1.client.__class__ == NonRecordingClient
E   AttributeError: 'PotelScope' object has no attribute 'client'

tests.test_scope test_get_current_scope_tags

Stack Traces | 0.001s run time

tests/test_scope.py:341: in test_get_current_scope_tags
    assert scope1.client.__class__ == NonRecordingClient
E   AttributeError: 'PotelScope' object has no attribute 'client'

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

[sl0thentr0py]

nice!