fix: don't adopt a foreign global OTel span as a Judgment parent

[arthi-arumugam-git]

Picks up the regression from #749 that #751 and #752 were both aimed at. #752 has been open since 6 July, so this is an independent take on it with a failing test.

The problem

#749 made get_current_context() read the global OTel context whenever this provider owns the global provider. That folded two different jobs into one method:

• write: publish the active Judgment span into the global context, so third-party instrumentation calling trace.get_current_span() can see it
• read, decide what a newly started Judgment span should be parented to

Only the write direction actually needs to be global. Reading parenting from the global context means whatever span a host application or unrelated instrumentation happens to have left current becomes the parent of the next Judgment span, which silently reroots it onto a foreign trace. Nothing errors, and the trace just shows up in the wrong tree with the wrong trace_id.

The change

Split the two directions.

• get_current_context() always reads Judgment's private runtime context. Parenting never consults the global context.
• get_global_context() added for the cases that genuinely want the global view.
• attach_context() attaches to the private context and, when we own the global provider, mirrors the same context into the global one. It returns both tokens in a _ContextToken so detach_context() can unwind both, global first.

The write path is unchanged in effect, so the interop #749 was after still works.

Tests

TestForeignGlobalParent adds two.

test_foreign_global_span_is_not_adopted_as_parent fails on main:

> assert span.parent is None
E assert SpanContext(trace_id=0x56a27dc4..., span_id=0x2832975e..., ...) is None

A foreign SDK span is made current in the global context, a Judgment root span is started, and on main that root comes back with a parent and the foreign trace's trace_id.

test_judgment_children_still_parent_to_judgment_spans is the guard in the other direction, with the same foreign span current, a nested Judgment span still parents to its Judgment parent and stays on the same trace.

The three existing TestGlobalContextBridge tests pass unchanged, including test_active_span_visible_to_external_when_global, which is the behaviour #749 was protecting. Full file is 19 passed.

Happy to adjust naming or fold this into #752 instead if you'd rather keep that one, I mostly wanted the failing test to exist somewhere.

---

[devin-ai-integration]

✅ Devin Review: No Issues Found

Devin Review analyzed this PR and found no potential bugs to report.

View in Devin Review to see 1 additional finding.

[arthi-arumugam-git]

Gentle ping: this has been unchanged and ready for review for 8 days. Anything needed from my side to move it forward?