Don't require the parent span to have a tracer in traced_from_parent_span()

[wbarnha]

Note: Before submitting this pull request, please review our contributing
guidelines.

Description

Rebased onto master now that #787 has landed. The scope of this PR has changed substantially — it opened as an alternative fix for #786, and is now just the one piece that #787 did not cover.

#787 landed the same root-cause fix while this was open, so this defers to it: faust/utils/_opentracing.py is taken from master verbatim, and its parity tests (test_opentracing_shim.py) stand as the guard for the stand-in itself. Nothing here re-litigates that.

What's left

traced_from_parent_span() still assumes the parent span it is handed has a tracer:

child = parent.tracer.start_span(...)

Master's fix means the stand-in always supplies one, so the crash reported in #786 is gone either way. But a span is whatever the configured tracer hands back, and parent.tracer is not guaranteed — opentracing.Span documents tracer as a property backed by whatever was passed to the constructor, and nothing stops it being None. Tracing is instrumentation; it shouldn't be what breaks a rebalance. So this runs the wrapped function untraced instead:

tracer = getattr(parent, "tracer", None)
if tracer is None:
    return fun(*args, **kwargs)

getattr covers both branches the old code had — "no parent at all" and "parent with no tracer" — so the if parent is not None test folds into it and the body de-indents. That accounts for most of the diff; the behavioural change is the two added lines.

Tests

Cut down to what test_opentracing_shim.py doesn't already cover, and moved to tests/unit/utils/test_tracing.py to match what they now exercise:

• A tracerless parent, passed as an argument and reached through the context variable.
• Error propagation through the traced wrapper, parametrized over traced and untraced parents — the untraced path is new code, and it still has to let the wrapped function's exception through.
• Both rebalance callbacks with the stand-in substituted for faust.utils.tracing.opentracing. test_opentracing_shim.py notes that it exercises the stand-in's objects directly and so "do[es] not prove the except ImportError fallback wiring itself" — these cover that wiring, and would have caught Kafka rebalancing fails when opentracing is not installed #786 end to end.

Three of the eight fail without the guard; the rebalance tests pass on master already and are here as regression cover.

Verification

• tests/unit + tests/functional: 2503 passed, 12 skipped.
• isort / black clean; flake8 output byte-identical to master (4 pre-existing hits, none in touched files).

Happy to close this instead if you'd rather leave traced_from_parent_span() strict — #786 itself is fixed on master either way.

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 96.20%. Comparing base (39ded3c) to head (d006fd2).

Additional details and impacted files

@@           Coverage Diff           @@
##           master     #788   +/-   ##
=======================================
  Coverage   96.20%   96.20%           
=======================================
  Files         110      110           
  Lines       11789    11789           
  Branches     1281     1281           
=======================================
  Hits        11342    11342           
  Misses        350      350           
  Partials       97       97           

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.