fix(observe): preserve context-manager errors and cleanup

[1fanwang]

What does this PR do?

An observed context-manager generator replaces the application's error with AttributeError and leaves its file open. This preserves the generator protocol under the tracing context, so exception handling and cleanup still run.

Interrupt and process-exit exceptions also finish their observations and reach the caller unchanged.

Type of change

• Bug fix
• New feature
• Breaking change
• Refactor
• Documentation update
• Tooling, CI, or repo maintenance

Verification

Testing Done

The reproducer is in the regression tests in this diff. It uses the real Langfuse client, an in-memory span exporter and real temporary files. It does not mock the tracing runtime or call a server or model.

Scenario	Before	After
Sync and async context-manager failures	Application error is masked; file stays open; no finished span	Application error is preserved; file closes; span finishes
The generator handles the error	Error handler is never reached	Error is suppressed; file closes; span finishes

The same tests ran against the fetched main commit with only the test file copied over, then against this branch.

Reproduction commands

set -e
uv sync --locked --python 3.12
CHECKOUT=$PWD
TEST=tests/unit/test_observe.py
BASELINE=$(mktemp -d)
git worktree add --detach "$BASELINE" 65392c73731e07711828745de337fdf7bba31fbc
cp "$TEST" "$BASELINE/$TEST"
set +e
(
  cd "$BASELINE" &&
  UV_PROJECT_ENVIRONMENT="$CHECKOUT/.venv" uv run --no-sync python -m pytest \
    "$TEST" -k context_manager_preserves_exception_handling \
    -n 0 -q --tb=short --log-cli-level=INFO
)
BASELINE_EXIT=$?
set -e
test "$BASELINE_EXIT" -eq 1
uv run --no-sync python -m pytest "$TEST" \
  -k context_manager_preserves_exception_handling \
  -n 0 -q --tb=short --log-cli-level=INFO

Before and after, prior to forced cleanup:

INFO     tests.unit.test_observe:test_observe.py:100 sync suppress=False error=AttributeError("'_ContextPreservedSyncGeneratorWrapper' object has no attribute 'throw'") resource_closed=False finished_spans=0
INFO     tests.unit.test_observe:test_observe.py:100 sync suppress=True error=AttributeError("'_ContextPreservedSyncGeneratorWrapper' object has no attribute 'throw'") resource_closed=False finished_spans=0
INFO     tests.unit.test_observe:test_observe.py:157 async suppress=False error=AttributeError("'_ContextPreservedAsyncGeneratorWrapper' object has no attribute 'athrow'") resource_closed=False finished_spans=0
INFO     tests.unit.test_observe:test_observe.py:157 async suppress=True error=AttributeError("'_ContextPreservedAsyncGeneratorWrapper' object has no attribute 'athrow'") resource_closed=False finished_spans=0
exit=1

After, prior to forced cleanup:
INFO     tests.unit.test_observe:test_observe.py:100 sync suppress=False error=ValueError('application failed') resource_closed=True finished_spans=1
INFO     tests.unit.test_observe:test_observe.py:100 sync suppress=True error=None resource_closed=True finished_spans=1
INFO     tests.unit.test_observe:test_observe.py:100 sync suppress=False error=KeyboardInterrupt('application failed') resource_closed=True finished_spans=1
INFO     tests.unit.test_observe:test_observe.py:100 sync suppress=False error=SystemExit('application failed') resource_closed=True finished_spans=1
INFO     tests.unit.test_observe:test_observe.py:157 async suppress=False error=ValueError('application failed') resource_closed=True finished_spans=1
INFO     tests.unit.test_observe:test_observe.py:157 async suppress=True error=None resource_closed=True finished_spans=1
INFO     tests.unit.test_observe:test_observe.py:157 async suppress=False error=KeyboardInterrupt('application failed') resource_closed=True finished_spans=1
INFO     tests.unit.test_observe:test_observe.py:157 async suppress=False error=SystemExit('application failed') resource_closed=True finished_spans=1
exit=0

Checklist

• I self-reviewed the diff using code_review.md.
• I added or updated tests for behavior changes.
• I updated docs, examples, or .env.template if needed.
• I did not hand-edit generated files; if generated files changed, I used the upstream regeneration path.
• I did not commit secrets or credentials.

[claude]

Claude Code Review

This pull request is from a fork — automated review is disabled. A repository maintainer can comment @claude review to run a one-time review.

[CLAassistant]

All committers have signed the CLA.